Showing posts with label Nix. Show all posts
Showing posts with label Nix. Show all posts

Tuesday, August 31, 2021

A more elaborate approach for bypassing NPM's dependency management features in Nix builds

Nix is a general purpose package manager that can be used to automate the deployments of a variety of systems -- it can deploy components written in a variety of programming languages (e.g. C, C++, Java, Go, Rust, Perl, Python, JavaScript) using various kinds of technologies and frameworks, such as Django, Android, and Node.js.

Another unique selling point of Nix is that it provides strong reproducibility guarantees. If a build succeeds on one machine, then performing the same build on another should result in a build that is (nearly) bit-identical.

Nix improves build reproducibility by complementing build processes with features, such as:

  • Storing all artifacts in isolation in a so-called Nix store: /nix/store (e.g. packages, configuration files), in which every path is unique by prefixing it with an SHA256 hash code derived from all build inputs (e.g. dependencies, build scripts etc.). Isolated paths make it possible for multiple variants and versions of the same packages to safely co-exist.
  • Clearing environment variables or setting them to dummy values. In combination with unique and isolated Nix store paths, search environment variables must configured in such a way that the build script can find its dependencies in the Nix store, or it will fail.

    Having to specify all search environment variables may sound inconvenient, but prevents undeclared dependencies to accidentally make a build succeed -- deployment of such a package is very likely to fail on machine that misses an unknown dependency.
  • Running builds as an unprivileged user that does not have any rights to make modifications to the host system -- a build can only write in its designated temp folder or output paths.
  • Optionally running builds in a chroot environment, so that a build cannot possibly find any undeclared host system dependencies through hard-coded absolute paths.
  • Restricting network access to prevent a build from obtaining unknown dependencies that may influence the build outcome.

For many build tools, the Nixpkgs repository provides abstraction functions that allow you to easily construct a package from source code (e.g. GNU Make, GNU Autotools, Apache Ant, Perl's MakeMaker, SCons etc.).

However, certain tools are difficult to use in combination with Nix -- for example, NPM that is used to deploy Node.js projects.

NPM is both a dependency and build manager and the former aspect conflicts with Nix -- builds in Nix are typically prevented from downloading files from remote network locations, with the exception of so-called fixed-output derivations in which the output hash is known in advance.

If network connections would be allowed in regular builds, then Nix can no longer ensure that a build is reproducible (i.e. that the hash code in the Nix store path reflects the same build output derived from all inputs).

To cope with the conflicting dependency management feature of NPM, various kinds of integrations have been developed. npm2nix was the first, and several years ago I have started node2nix to provide a solution that aims for accuracy.

Basically, the build process of an NPM package in Nix boils down to performing the following steps in a Nix derivation:

# populate the node_modules/ folder
npm install --offline

We must first obtain the required dependencies of a project through the Nix package manager and install them in the correct locations in the node_modules/ directory tree.

Finally, we should run NPM in offline mode forcing it not to re-obtain or re-install any dependencies, but still perform build management tasks, such as running build scripts.

From a high-level point of view, this principle may look simple, but in practice it is not:

  • With earlier versions of NPM, we were forced to imitate its dependency resolution algorithm. At first sight, it looked simple, but getting it right (such as coping with circular dependencies and dependency de-duplication) is much more difficult than expected.
  • NPM 5.x introduced lock files. For NPM development projects, they provide exact version specifiers of all dependencies and transitive dependencies, making it much easier to know which dependencies need to be installed.

    Unfortunately, NPM also introduced an offline cache, that prevents us from simply copying packages into the node_modules/ tree. As a result, we need to make additional complex modifications to the package.json configuration files of all dependencies.

    Furthermore, end user package installations do not work with lock files, requiring us to still keep our custom implementation of the dependency resolution algorithm.
  • NPM's behaviour with dependencies on directories on the local file system has changed. In old versions of NPM, such dependencies were copied, but in newer versions, they are symlinked. Furthermore, each directory dependency maintains its own node_modules/ directory for transitive dependencies.

Because we need to take many kinds of installation scenarios into account and work around the directory dependency challenges, the implementation of the build environment: node-env.nix in node2nix has become very complicated.

It has become so complicated that I consider it a major impediment in making any significant changes to the build environment.

In the last few weeks, I have been working on a companion tool named: placebo-npm that should simplify the installation process. Moreover, it should also fix a number of frequently reported issues.

In this blog post, I will explain how the tool works.

Lock-driven deployments


In NPM 5.x, package-lock.json files were introduced. The fact that they capture the exact versions of all dependencies and make all transitive dependencies known, makes certain aspects of an NPM deployment in a Nix build environment easier.

For lock-driven projects, we no longer have to run our own implementation of the dependency resolution algorithm to figure out what the exact versions of all dependencies and transitive dependencies are.

For example, a project with the following package.json:

{
  "name": "simpleproject",
  "version": "0.0.1",
  "dependencies": {
    "underscore": "*",
    "prom2cb": "github:svanderburg/prom2cb",
    "async": "https://mylocalserver/async-3.2.1.tgz"
  }
}

may have the following package-lock.json file:

{
  "name": "simpleproject",
  "version": "0.0.1",
  "lockfileVersion": 1,
  "requires": true,
  "dependencies": {
    "async": {
      "version": "https://mylocalserver/async-3.2.1.tgz",
      "integrity": "sha512-XdD5lRO/87udXCMC9meWdYiR+Nq6ZjUfXidViUZGu2F1MO4T3XwZ1et0hb2++BgLfhyJwy44BGB/yx80ABx8hg=="
    },
    "prom2cb": {
      "version": "github:svanderburg/prom2cb#fab277adce1af3bc685f06fa1e43d889362a0e34",
      "from": "github:svanderburg/prom2cb"
    },
    "underscore": {
      "version": "1.13.1",
      "resolved": "https://registry.npmjs.org/underscore/-/underscore-1.13.1.tgz",
      "integrity": "sha512-hzSoAVtJF+3ZtiFX0VgfFPHEDRm7Y/QPjGyNo4TVdnDTdft3tr8hEkD25a1jC+TjTuE7tkHGKkhwCgs9dgBB2g=="
    }
  }
}

As you may notice, the package.json file declares three dependencies:

  • The first dependency is underscore that refers to the latest version in the NPM registry. In the package-lock.json file, the dependency is frozen to version 1.13.1. The resolved property provides the URL where the tarball should be obtained from. Its integrity can be verified with the given SHA512 hash.
  • The second dependency: prom2cb refers to the latest revision of the main branch of the prom2cb Git repository on GitHub. In the package-lock.json file, it is pinpointed to the fab277... revision.
  • The third dependency: async refers to a tarball that is downloaded from an arbitrary HTTP URL. The package-lock.json records its SHA512 integrity hash to make sure that we can only deploy with the version that we have used previously.

As explained earlier, to ensure purity, in a Nix build environment, we cannot allow NPM to obtain the required dependencies of a project. Instead, we must let Nix obtain all the dependencies.

When all dependencies have been obtained, we should populate the node_modules/ folder of the project. In the above example, it is just simply a matter of unpacking the tarballs or copying the Git clones into the node_modules/ folder of the project. No transitive dependencies need to be deployed.

For projects that do not rely on build scripts (that perform tasks, such as linting, compiling code, such as TypeScript etc.) this typically suffices to make a project work.

However, when we also need build management, we need to run the full installation process:

$ npm install --offline

npm ERR! code ENOTCACHED
npm ERR! request to https://registry.npmjs.org/async/-/async-3.2.1.tgz failed: cache mode is 'only-if-cached' but no cached response available.

npm ERR! A complete log of this run can be found in:
npm ERR!     /home/sander/.npm/_logs/2021-08-29T12_56_13_978Z-debug.log

Unfortunately, NPM still tries to obtain the dependencies despite the fact that they have already been copied into the right locations into node_modules folder.

Bypassing the offline cache


To cope with the problem that manually obtained dependencies cannot be detected, my initial idea was to use the NPM offline cache in a specific way.

The offline cache claims to be content-addressable, meaning that every item can be looked up by using a hash code that represents its contents, regardless of its origins. Unfortunately, it turns out that this property cannot be fully exploited.

For example, when we obtain the underscore tarball (with the exact same contents) from a different URL:

$ npm cache add http://mylocalcache/underscore-1.13.1.tgz

and run the installation in offline mode:

$ npm install --offline
npm ERR! code ENOTCACHED
npm ERR! request to https://registry.npmjs.org/underscore/-/underscore-1.13.1.tgz failed: cache mode is 'only-if-cached' but no cached response available.

npm ERR! A complete log of this run can be found in:
npm ERR!     /home/sander/.npm/_logs/2021-08-26T13_50_15_137Z-debug.log

The installation still fails, despite the fact that we already have a tarball (with the exact same SHA512 hash) in our cache.

However, downloading underscore from its original location (the NPM registry):

$ npm cache add underscore@1.13.1

makes the installation succeed.

The reason why downloading the same tarball from an arbitrary HTTP URL does not work is because NPM will only compute a SHA1 hash. Obtaining a tarball from the NPM registry causes NPM to compute a SHA512 hash. Because it was downloaded from a different source, it fails to recognize the SHA512 hash in the package-lock.json file.

We also run into similar issues when we obtain an old package from the NPM registry that only has an SHA1 hash. Importing the same file from a local file path causes NPM to compute a SHA512 hash. As a result, npm install tries to re-obtain the same tarball from the remote location, because the hash was not recognized.

To cope with these problems, placebo-npm will completely bypass the cache. After all dependencies have been copied to the node_modules folder, it modifies their package.json configuration files with hidden metadata properties to trick NPM that they came from their original locations.

For example, to make the underscore dependency work (that is normally obtained from the NPM registry), we must add the following properties to the package.json file:

{
  ...
  _from: "underscore@https://registry.npmjs.org/underscore/-/underscore-1.13.1.tgz",
  _integrity: "sha512-XdD5lRO/87udXCMC9meWdYiR+Nq6ZjUfXidViUZGu2F1MO4T3XwZ1et0hb2++BgLfhyJwy44BGB/yx80ABx8hg==",
  _resolved: "https://registry.npmjs.org/underscore/-/underscore-1.13.1.tgz"
}

For prom2cb (that is a Git dependency), we should add:

{
  ...
  _from = "github:svanderburg/prom2cb",
  _integrity = "",
  _resolved = "github:svanderburg/prom2cb#fab277adce1af3bc685f06fa1e43d889362a0e34"
}

and for HTTP/HTTPS dependencies and local files we should do something similar (adding _from and _integrity fields).

With these modifications, NPM will no longer attempt to consult the local cache, making the dependency installation step succeed.

Handling directory dependencies


Another challenge is dependencies on local directories, that are frequently used for local development projects:

{
  "name": "simpleproject",
  "version": "0.0.1",
  "dependencies": {
    "underscore": "*",
    "prom2cb": "github:svanderburg/prom2cb",
    "async": "https://mylocalserver/async-3.2.1.tgz",
    "mydep": "../../mydep",
  }
}

In the package.json file shown above, a new dependency has been added: mydep that refers to a relative local directory dependency: ../../mydep.

If we run npm install, then NPM creates a symlink to the folder in the project's node_modules/ folder and installs the transitive dependencies in the node_modules/ folder of the target dependency.

If we want to deploy the same project to a different machine, then it is required to put mydep in the exact same relative location, or the deployment will fail.

Deploying such an NPM project with Nix introduces a new problem -- all packages deployed by Nix are stored in the Nix store (typically /nix/store). After deploying the project, the relative path to the project (from the Nix store) will no longer be correct. Moreover, we also want Nix to automatically deploy the directory dependency as part of the deployment of the entire project.

To cope with these inconveniences, we are required to implement a tricky solution -- we must rewrite directory dependencies in such a way that can refer to a folder that is automatically deployed by Nix. Furthermore, the dependency should still end up being symlink to satisfy NPM -- copying directory dependencies in the node_modules/ folder is not accepted by NPM.

Usage


To conveniently install NPM dependencies from a local source (and satisfying npm in such a way that it believes the dependencies came from their original locations), I have created a tool called: placebo-npm.

We can, for example, obtain all required dependencies ourselves and put them in a local cache folder:

$ mkdir /home/sander/mycache
$ wget https://mylocalserver/async-3.2.1.tgz
$ wget https://registry.npmjs.org/underscore/-/underscore-1.13.1.tgz
$ git clone https://github.com/svanderburg/prom2cb

The deployment process that placebo-npm executes is driven by a package-placebo.json configuration file that has the following structure:

{
   "integrityHashToFile": {
     "sha512-hzSoAVtJF+3ZtiFX0VgfFPHEDRm7Y/QPjGyNo4TVdnDTdft3tr8hEkD25a1jC+TjTuE7tkHGKkhwCgs9dgBB2g==": "/home/sander/mycache/underscore-1.13.1.tgz",
     "sha512-XdD5lRO/87udXCMC9meWdYiR+Nq6ZjUfXidViUZGu2F1MO4T3XwZ1et0hb2++BgLfhyJwy44BGB/yx80ABx8hg==": "/home/sander/mycache/async-3.2.1.tgz"
   },
   "versionToFile": {
     github:svanderburg/prom2cb#fab277adce1af3bc685f06fa1e43d889362a0e34": "/home/sander/mycache/prom2cb"
   },
   "versionToDirectoryCopyLink": {
     "file:../dep": "/home/sander/alternatedir/dep"
   }
}

The placebo config maps dependencies in a package-lock.json file to local file references:

  • integrityHashToFile maps dependencies with an integrity hash to local files, which is useful for HTTP/HTTPS dependencies, registry dependencies, and local file dependencies.
  • versionToFile: maps dependencies with a version property to local directories. This is useful for Git dependencies.
  • versionToDirectoryCopyLink: specifies directories that need to be copied into a shadow directory named: placebo_node_dirs and creates symlinks to the shadow directories in the node_modules/ folder. This is useful for installing directory dependencies from arbitrary locations.

With the following command, we can install all required dependencies from the local cache directory and make all necessary modifications to let NPM accept the dependencies:

$ placebo-npm package-placebo.json

Finally, we can run:

$ npm install --offline

The above command does not attempt to re-obtain or re-install the dependencies, but still performs all required build management tasks.

Integration with Nix


All the functionality that placebo-npm provides has already been implemented in the node-env.nix module, but over the years it has evolved into a very complex beast -- it is implemented as a series of Nix functions that generates shell code.

As a consequence, it suffers from recursion problems and makes it extremely difficult to tweak/adjust build processes, such as modifying environment variables or injecting arbitrary build steps to work around Nix integration problems.

With placebo-npm we can reduce the Nix expression that builds projects (buildNPMProject) to an implementation that roughly has the following structure:

{stdenv, placebo-npm}:
{packagePlacebo}:

stdenv.mkDerivation ({
  pname = builtins.replaceStrings [ "@" "/" ] [ "_at_" "_slash_" ] pname; # Escape characters that aren't allowed in a store path

  placeboJSON = builtins.toJSON packagePlacebo;
  passAsFile = [ "placeboJSON" ];

  buildInputs = [ nodejs placebo-npm ] ++ buildInputs;

  buildPhase = ''
    runHook preBuild
    true
    runHook postBuild
  '';
  installPhase = ''
    runHook preInstall

    mkdir -p $out/lib/node_modules/${pname}
    mv * $out/lib/node_modules/${pname}
    cd $out/lib/node_modules/${pname}

    placebo-npm --placebo $placeboJSONPath
    npm install --offline

    runHook postInstall
  '';
} // extraArgs)

As may be observed, the implementation is much more compact and fits easily on one screen. The function accepts a packagePlacebo attribute set as a parameter (that gets translated into a JSON file by the Nix package manager).

Aside from some simple house keeping work, most of the complex work has been delegated to executing placebo-npm inside the build environment, before we run npm install.

The function above is also tweakable -- it is possible to inject arbitrary environment variables and adjust the build process through build hooks (e.g. preInstall and postInstall).

Another bonus feature of delegating all dependency installation functionality to the placebo-npm tool is that we can also use this tool as a build input for other kinds projects -- we can use it the construction process of systems that are built from monolithic repositories, in which NPM is invoked from the build process of the encapsulating project.

The only requirement is to run placebo-npm before npm install is invoked.

Other use cases


In addition to using placebo-npm as a companion tool for node2nix and setting up a simple local cache, it can also be useful to facilitate offline installations from external media, such as USB flash drives.

Discussion


With placebo-npm we can considerably simplify the implementation of node-env.nix (part of node2nix) making it much easier to maintain. I consider the node-env.nix module the second most complicated aspect of node2nix.

As a side effect, it has also become quite easy to provide tweakable build environments -- this should solve a large number of reported issues. Many reported issues are caused by the fact that it is difficult or sometimes impossible to make changes to a project so that it will cleanly deploy.

Moreover, placebo-npm can also be used as a build input for projects built from monolithic repositories, in which a sub set needs to be deployed by NPM.

The integration of the new node-env.nix implementation into node2nix is not completely done yet. I have reworked it, but the part that generates the package-placebo.json file and lets Nix obtain all required dependencies is still a work-in-progress.

I am experimenting with two implementations: a static approach that generates Nix expressions and dynamic implementation that directly consumes a package-lock.json file in the Nix expression language. Both approaches have pros and cons. As a result, node2nix needs to combine both of them into a hybrid approach.

In a next blog post, I will explain more about them.

Availability


The initial version of placebo-npm can be obtained from my GitHub page.

Monday, April 26, 2021

A test framework for the Nix process management framework

As already explained in many previous blog posts, the Nix process management framework adds new ideas to earlier service management concepts explored in Nixpkgs and NixOS:

  • It makes it possible to deploy services on any operating system that can work with the Nix package manager, including conventional Linux distributions, macOS and FreeBSD. It also works on NixOS, but NixOS is not a requirement.
  • It allows you to construct multiple instances of the same service, by using constructor functions that identify conflicting configuration parameters. These constructor functions can be invoked in such a way that these configuration properties no longer conflict.
  • We can target multiple process managers from the same high-level deployment specifications. These high-level specifications are automatically translated to parameters for a target-specific configuration function for a specific process manager.

    It is also possible to override or augment the generated parameters, to work with configuration properties that are not universally supported.
  • There is a configuration option that conveniently allows you to disable user changes making it possible to deploy services as an unprivileged user.

Although the above features are interesting, one particular challenge is that the framework cannot guarantee that all possible variations will work after writing a high-level process configuration. The framework facilitates code reuse, but it is not a write once, run anywhere approach.

To make it possible to validate multiple service variants, I have developed a test framework that is built on top of the NixOS test driver that makes it possible to deploy and test a network of NixOS QEMU virtual machines with very minimal storage and RAM overhead.

In this blog post, I will describe how the test framework can be used.

Automating tests


Before developing the test framework, I was mostly testing all my packaged services manually. Because a manual test process is tedious and time consuming, I did not have any test coverage for anything but the most trivial example services. As a result, I frequently ran into many configuration breakages.

Typically, when I want to test a process instance, or a system that is composed of multiple collaborative processes, I perform the following steps:

  • First, I need to deploy the system for a specific process manager and configuration profile, e.g. for a privileged or unprivileged user, in an isolated environment, such as a virtual machine or container.
  • Then I need to wait for all process instances to become available. Readiness checks are critical and typically more complicated than expected -- for most services, there is a time window between a successful invocation of a process and its availability to carry out its primary task, such as accepting network connections. Executing tests before a service is ready, typically results in errors.

    Although there are process managers that can generally deal with this problem (e.g. systemd has the sd_notify protocol and s6 its own protocol and a sd_notify wrapper), the lack of a standardized protocol and its adoption still requires me to manually implement readiness checks.

    (As a sidenote: the only readiness check protocol that is standardized is for traditional System V services that daemonize on their own. The calling parent process should almost terminate immediately, but still wait until the spawned daemon child process notifies it to be ready.

    As described in an earlier blog post, this notification aspect is more complicated to implement than I thought. Moreover, not all traditional System V daemons follow this protocol.)
  • When all process instances are ready, I can check whether they properly carry out their tasks, and whether the integration of these processes work as expected.

An example


I have developed a Nix function: testService that automates the above process using the NixOS test driver -- I can use this function to create a test suite for systems that are made out of running processes, such as the webapps example described in my previous blog posts about the Nix process management framework.

The example system consists of a number of webapp processes with an embedded HTTP server returning HTML pages displaying their identities. Nginx reverse proxies forward incoming connections to the appropriate webapp processes by using their corresponding virtual host header values:

{ pkgs ? import <nixpkgs> { inherit system; }
, system ? builtins.currentSystem
, stateDir ? "/var"
, runtimeDir ? "${stateDir}/run"
, logDir ? "${stateDir}/log"
, cacheDir ? "${stateDir}/cache"
, libDir ? "${stateDir}/lib"
, tmpDir ? (if stateDir == "/var" then "/tmp" else "${stateDir}/tmp")
, forceDisableUserChange ? false
, processManager
}:

let
  sharedConstructors = import ../../../examples/services-agnostic/constructors/constructors.nix {
    inherit pkgs stateDir runtimeDir logDir cacheDir libDir tmpDir forceDisableUserChange processManager;
  };

  constructors = import ../../../examples/webapps-agnostic/constructors/constructors.nix {
    inherit pkgs stateDir runtimeDir logDir tmpDir forceDisableUserChange processManager;
    webappMode = null;
  };
in
rec {
  webapp1 = rec {
    port = 5000;
    dnsName = "webapp1.local";

    pkg = constructors.webapp {
      inherit port;
      instanceSuffix = "1";
    };
  };

  webapp2 = rec {
    port = 5001;
    dnsName = "webapp2.local";

    pkg = constructors.webapp {
      inherit port;
      instanceSuffix = "2";
    };
  };

  webapp3 = rec {
    port = 5002;
    dnsName = "webapp3.local";

    pkg = constructors.webapp {
      inherit port;
      instanceSuffix = "3";
    };
  };

  webapp4 = rec {
    port = 5003;
    dnsName = "webapp4.local";

    pkg = constructors.webapp {
      inherit port;
      instanceSuffix = "4";
    };
  };

  nginx = rec {
    port = if forceDisableUserChange then 8080 else 80;
    webapps = [ webapp1 webapp2 webapp3 webapp4 ];

    pkg = sharedConstructors.nginxReverseProxyHostBased {
      inherit port webapps;
    } {};
  };

  webapp5 = rec {
    port = 5004;
    dnsName = "webapp5.local";

    pkg = constructors.webapp {
      inherit port;
      instanceSuffix = "5";
    };
  };

  webapp6 = rec {
    port = 5005;
    dnsName = "webapp6.local";

    pkg = constructors.webapp {
      inherit port;
      instanceSuffix = "6";
    };
  };

  nginx2 = rec {
    port = if forceDisableUserChange then 8081 else 81;
    webapps = [ webapp5 webapp6 ];

    pkg = sharedConstructors.nginxReverseProxyHostBased {
      inherit port webapps;
      instanceSuffix = "2";
    } {};
  };
}

The processes model shown above (processes-advanced.nix) defines the following process instances:

  • There are six webapp process instances, each running an embedded HTTP service, returning HTML pages with their identities. The dnsName property specifies the DNS domain name value that should be used as a virtual host header to make the forwarding from the reverse proxies work.
  • There are two nginx reverse proxy instances. The former: nginx forwards incoming connections to the first four webapp instances. The latter: nginx2 forwards incoming connections to webapp5 and webapp6.

With the following command, I can connect to webapp2 through the first nginx reverse proxy:

$ curl -H 'Host: webapp2.local' http://localhost:8080
<!DOCTYPE html>
<html>
  <head>
    <title>Simple test webapp</title>
  </head>
  <body>
    Simple test webapp listening on port: 5001
  </body>
</html>

Creating a test suite


I can create a test suite for the web application system as follows:

{ pkgs, testService, processManagers, profiles }:

testService {
  exprFile = ./processes.nix;

  readiness = {instanceName, instance, ...}:
    ''
      machine.wait_for_open_port(${toString instance.port})
    '';

  tests = {instanceName, instance, ...}:
    pkgs.lib.optionalString (instanceName == "nginx" || instanceName == "nginx2")
      (pkgs.lib.concatMapStrings (webapp: ''
        machine.succeed(
            "curl --fail -H 'Host: ${webapp.dnsName}' http://localhost:${toString instance.port} | grep ': ${toString webapp.port}'"
        )
      '') instance.webapps);

  inherit processManagers profiles;
}

The Nix expression above invokes testService with the following parameters:

  • processManagers refers to a list of names of all the process managers that should be tested.
  • profiles refers to a list of configuration profiles that should be tested. Currently, it supports privileged for privileged deployments, and unprivileged for unprivileged deployments in an unprivileged user's home directory, without changing user permissions.
  • The exprFile parameter refers to the processes model of the system: processes-advanced.nix shown earlier.
  • The readiness parameter refers to a function that does a readiness check for each process instance. In the above example, it checks whether each service is actually listening on the required TCP port.
  • The tests parameter refers to a function that executes tests for each process instance. In the above example, it ignores all but the nginx instances, because explicitly testing a webapp instance is a redundant operation.

    For each nginx instance, it checks whether all webapp instances can be reached from it, by running the curl command.

The readiness and tests functions take the following parameters: instanceName identifies the process instance in the processes model, and instance refers to the attribute set containing its configuration.

Furthermore, they can refer to global process model configuration parameters:

  • stateDir: The directory in which state files are stored (typically /var for privileged deployments)
  • runtimeDir: The directory in which runtime files are stored (typically /var/run for privileged deployments).
  • forceDisableUserChange: Indicates whether to disable user changes (for unprivileged deployments) or not.

In addition to writing tests that work on instance level, it is also possible to write tests on system level, with the following parameters (not shown in the example):

  • initialTests: instructions that run right after deploying the system, but before the readiness checks, and instance-level tests.
  • postTests: instructions that run after the instance-level tests.

The above functions also accept the same global configuration parameters, and processes that refers to the entire processes model.

We can also configure other properties useful for testing:

  • systemPackages: installs additional packages into the system profile of the test virtual machine.
  • nixosConfig defines a NixOS module with configuration properties that will be added to the NixOS configuration of the test machine.
  • extraParams propagates additional parameters to the processes model.

Composing test functions


The Nix expression above is not self-contained. It is a function definition that needs to be invoked with all required parameters including all the process managers and profiles that we want to test for.

We can compose tests in the following Nix expression:

{ pkgs ? import <nixpkgs> { inherit system; }
, system ? builtins.currentSystem
, processManagers ? [ "supervisord" "sysvinit" "systemd" "disnix" "s6-rc" ]
, profiles ? [ "privileged" "unprivileged" ]
}:

let
  testService = import ../../nixproc/test-driver/universal.nix {
    inherit system;
  };
in
{

  nginx-reverse-proxy-hostbased = import ./nginx-reverse-proxy-hostbased {
    inherit pkgs processManagers profiles testService;
  };

  docker = import ./docker {
    inherit pkgs processManagers profiles testService;
  };

  ...
}

The above partial Nix expression (default.nix) invokes the function defined in the previous Nix expression that resides in the nginx-reverse-proxy-hostbased directory and propagates all required parameters. It also composes other test cases, such as docker.

The parameters of the composition expression allow you to globally configure all the desired service variants:

  • processManagers allows you to select the process managers you want to test for.
  • profiles allows you to select the configuration profiles.

With the following command, we can test our system as a privileged user, using systemd as a process manager:

$ nix-build -A nginx-reverse-proxy-hostbased.privileged.systemd

we can also run the same test, but then as an unprivileged user:

$ nix-build -A nginx-reverse-proxy-hostbased.unprivileged.systemd

In addition to systemd, any configured process manager can be used that works in NixOS. The following command runs a privileged test of the same service for sysvinit:

$ nix-build -A nginx-reverse-proxy-hostbased.privileged.sysvinit

Results


With the test driver in place, I have managed to expand my repository of example services, provided test coverage for them and fixed quite a few bugs in the framework caused by regressions.

Below is a screenshot of Hydra: the Nix-based continuous integration service showing an overview of test results for all kinds of variants of a service:


So far, the following services work multi-instance, with multiple process managers, and (optionally) as an unprivileged user:

  • Apache HTTP server. In the services repository, there are multiple constructors for deploying an Apache HTTP server: to deploy static web applications or dynamic web applications with PHP, and to use it as a reverse proxy (via HTTP and AJP) with HTTP basic authentication optionally enabled.
  • Apache Tomcat.
  • Nginx. For Nginx we also have multiple constructors. One to deploy a configuration for serving static web apps, and two for setting up reverse proxies using paths or virtual hosts to forward incoming requests to the appropriate services.

    The reverse proxy constructors can also generate configurations that will cache the responses of incoming requests.
  • MySQL/MariaDB.
  • PostgreSQL.
  • InfluxDB.
  • MongoDB.
  • OpenSSH.
  • svnserve.
  • xinetd.
  • fcron. By default, the fcron user and group are hardwired into the executable. To facilitate unprivileged user deployments, we automatically create a package build override to propagate the --with-run-non-privileged configuration flag so that it can run as unprivileged user. Similarly, for multiple instances we create an override to use a different user and group that does not conflict with the primary instance.
  • supervisord
  • s6-svscan

The following service also works with multiple instances and multiple process managers, but not as an unprivileged user:


The following services work with multiple process managers, but not multi-instance or as an unprivileged user:

  • D-Bus
  • Disnix
  • nix-daemon
  • Hydra

In theory, the above services could be adjusted to work as an unprivileged user, but doing so is not very useful -- for example, the nix-daemon's purpose is to facilitate multi-user package deployments. As an unprivileged user, you only want to facilitate package deployments for yourself.

Moreover, the multi-instance aspect is IMO also not very useful to explore for these services. For example, I can not think of a useful scenario to have two Hydra instances running next to each other.

Discussion


The test framework described in this blog post is an important feature addition to the Nix process management framework -- it allowed me to package more services and fix quite a few bugs caused by regressions.

I can now finally show that it is doable to package services and make them work under nearly all possible conditions that the framework supports (e.g. multiple instances, multiple process managers, and unprivileged user installations).

The only limitation of the test framework is that it is not operating system agnostic -- the NixOS test driver (that serves as its foundation), only works (as its name implies) with NixOS, which itself is a Linux distribution. As a result, we can not automatically test bsdrc scripts, launchd daemons, and cygrunsrv services.

In theory, it is also possible to make a more generalized test driver that works with multiple operating systems. The NixOS test driver is a combination of ideas (e.g. a shared Nix store between the host and guest system, an API to control QEMU, and an API to manage services). We could also dissect these ideas and run them on conventional QEMU VMs running different operating systems (with the Nix package manager).

Although making a more generalized test driver is interesting, it is beyond the scope of the Nix process management framework (which is about managing process instances, not entire systems).

Another drawback is that while it is possible to test all possible service variants on Linux, it may be very expensive to do so.

However, full process manager coverage is often not required to get a reasonable level of confidence. For many services, it typically suffices to implement the following strategy:

  • Pick two process managers: one that prefers foreground processes (e.g. supervisord) and one that prefers daemons (e.g. sysvinit). This is the most significant difference (from a configuration perspective) between all these different process managers.
  • If a service supports multiple configuration variants, and multiple instances, then create a processes model that concurrently deploys all these variants.

Implementing the above strategy only requires you to test four variants, providing a high degree of certainty that it will work with all other process managers as well.

Future work


Most of the interesting functionality required to work with the Nix process management framework is now implemented. I still need to implement more changes to make it more robust and "dog food" more of my own problems as much as possible.

Moreover, the docker backend still requires a bit more work to make it more usable.

Eventually, I will be thinking of an RFC that will upstream the interesting bits of the framework into Nixpkgs.

Availability


The Nix process management framework repository as well as the example services repository can be obtained from my GitHub page.

Friday, March 12, 2021

Using the Nix process management framework as an infrastructure deployment solution for Disnix

As explained in many previous blog posts, I have developed Disnix as a solution for automating the deployment of service-oriented systems -- it deploys heterogeneous systems, that consist of many different kinds of components (such as web applications, web services, databases and processes) to networks of machines.

The deployment models for Disnix are typically not fully self-contained. Foremost, a precondition that must be met before a service-oriented system can be deployed, is that all target machines in the network require the presence of Nix package manager, Disnix, and a remote connectivity service (e.g. SSH).

For multi-user Disnix installations, in which the user does not have super-user privileges, the Disnix service is required to carry out deployment operations on behalf of a user.

Moreover, the services in the services model typically need to be managed by other services, called containers in Disnix terminology (not to be confused with Linux containers).

Examples of container services are:

  • The MySQL DBMS container can manage multiple databases deployed by Disnix.
  • The Apache Tomcat servlet container can manage multiple Java web applications deployed by Disnix.
  • systemd can act as a container that manages multiple systemd units deployed by Disnix.

Managing the life-cycles of services in containers (such as activating or deactivating them) is done by a companion tool called Dysnomia.

In addition to Disnix, these container services also typically need to be deployed in advance to the target machines in the network.

The problem domain that Disnix works in is called service deployment, whereas the deployment of machines (bare metal or virtual machines) and the container services is called infrastructure deployment.

Disnix can be complemented with a variety of infrastructure deployment solutions:

  • NixOps can deploy networks of NixOS machines, both physical and virtual machines (in the cloud), such as Amazon EC2.

    As part of a NixOS configuration, the Disnix service can be deployed that facilitates multi-user installations. The Dysnomia NixOS module can expose all relevant container services installed by NixOS as container deployment targets.
  • disnixos-deploy-network is a tool that is included with the DisnixOS extension toolset. Since services in Disnix can be any kind of deployment unit, it is also possible to deploy an entire NixOS configuration as a service. This tool is mostly developed for demonstration purposes.

    A limitation of this tool is that it cannot instantiate virtual machines and bootstrap Disnix.
  • Disnix itself. The above solutions are all NixOS-based, a software distribution that is Linux-based and fully managed by the Nix package manager.

    Although NixOS is very powerful, it has two drawbacks for Disnix:

    • NixOS uses the NixOS module system for configuring system aspects. It is very powerful but you can only deploy one instance of a system service -- Disnix can also work with multiple container instances of the same type on a machine.
    • Services in NixOS cannot be deployed to other kinds software distributions: conventional Linux distributions, and other operating systems, such as macOS and FreeBSD.

    To overcome these limitations, Disnix can also be used as a container deployment solution on any operating system that is capable of running Nix and Disnix. Services deployed by Disnix can automatically be exposed as container providers.

    Similar to disnix-deploy-network, a limitation of this approach is that it cannot be used to bootstrap Disnix.

Last year, I have also added a new major feature to Disnix making it possible to deploy both application and container services in the same Disnix deployment models, minimizing the infrastructure deployment problem -- the only requirement is to have machines with Nix, Disnix, and a remote connectivity service (such as SSH) pre-installed on them.

Although this integrated feature is quite convenient, in particular for test setups, a separated infrastructure deployment process (that includes container services) still makes sense in many scenarios:

  • The infrastructure parts and service parts can be managed by different people with different specializations. For example, configuring and tuning an application server is a different responsibility than developing a Java web application.
  • The service parts typically change more frequently than the infrastructure parts. As a result, they typically have different kinds of update cycles.
  • The infrastructure components can typically be reused between projects (e.g. many systems use a database backend such as PostgreSQL or MySQL), whereas the service components are typically very project specific.

I also realized that my other project: the Nix process management framework can serve as a partial infrastructure deployment solution -- it can be used to bootstrap Disnix and deploy container services.

Moreover, it can also deploy multiple instances of container services and used on any operating system that the Nix process management framework supports, including conventional Linux distributions and other operating systems, such as macOS and FreeBSD.

Deploying and exposing the Disnix service with the Nix process management framework


As explained earlier, to allow Disnix to deploy services to a remote machine, a machine needs to have Disnix installed (and run the Disnix service for a multi-user installation), and be remotely connectible, e.g. through SSH.

I have packaged all required services as constructor functions for the Nix process management framework.

The following process model captures the configuration of a basic multi-user Disnix installation:

{ pkgs ? import <nixpkgs> { inherit system; }
, system ? builtins.currentSystem
, stateDir ? "/var"
, runtimeDir ? "${stateDir}/run"
, logDir ? "${stateDir}/log"
, spoolDir ? "${stateDir}/spool"
, cacheDir ? "${stateDir}/cache"
, tmpDir ? (if stateDir == "/var" then "/tmp" else "${stateDir}/tmp")
, forceDisableUserChange ? false
, processManager
}:

let
  ids = if builtins.pathExists ./ids-bare.nix then (import ./ids-bare.nix).ids else {};

  constructors = import ../../services-agnostic/constructors.nix {
    inherit pkgs stateDir runtimeDir logDir tmpDir cacheDir spoolDir forceDisableUserChange processManager ids;
  };
in
rec {
  sshd = {
    pkg = constructors.sshd {
      extraSSHDConfig = ''
        UsePAM yes
      '';
    };

    requiresUniqueIdsFor = [ "uids" "gids" ];
  };

  dbus-daemon = {
    pkg = constructors.dbus-daemon {
      services = [ disnix-service ];
    };

    requiresUniqueIdsFor = [ "uids" "gids" ];
  };

  disnix-service = {
    pkg = constructors.disnix-service {
      inherit dbus-daemon;
    };

    requiresUniqueIdsFor = [ "gids" ];
  };
}

The above processes model (processes.nix) captures three process instances:

  • sshd is the OpenSSH server that makes it possible to remotely connect to the machine by using the SSH protocol.
  • dbus-daemon runs a D-Bus system daemon, that is a requirement for the Disnix service. The disnix-service is propagated as a parameter, so that its service directory gets added to the D-Bus system daemon configuration.
  • disnix-service is a service that executes deployment operations on behalf of an authorized unprivileged user. The disnix-service has a dependency on the dbus-service making sure that the latter gets activated first.

We can deploy the above configuration on a machine that has the Nix process management framework already installed.

For example, to deploy the configuration on a machine that uses supervisord, we can run:

$ nixproc-supervisord-switch processes.nix

Resulting in a system that consists of the following running processes:

$ supervisorctl 
dbus-daemon                      RUNNING   pid 2374, uptime 0:00:34
disnix-service                   RUNNING   pid 2397, uptime 0:00:33
sshd                             RUNNING   pid 2375, uptime 0:00:34

As may be noticed, the above supervised services correspond to the processes in the processes model.

On the coordinator machine, we can write a bootstrap infrastructure model (infra-bootstrap.nix) that only contains connectivity settings:

{
  test1.properties.hostname = "192.168.2.1";
}

and use the bootstrap model to capture the full infrastructure model of the system:

$ disnix-capture-infra infra-bootstrap.nix

resulting in the following configuration:

{
  "test1" = {
    properties = {
      "hostname" = "192.168.2.1";
      "system" = "x86_64-linux";
    };
    containers = {
      echo = {
      };
      fileset = {
      };
      process = {
      };
      supervisord-program = {
        "supervisordTargetDir" = "/etc/supervisor/conf.d";
      };
      wrapper = {
      };
    };
    "system" = "x86_64-linux";
  };
}

Despite the fact that we have not configured any containers explicitly, the above configuration (infrastructure.nix) already exposes a number of container services:

  • The echo, fileset and process container services are built-in container providers that any Dysnomia installation includes.

    The process container can be used to automatically deploy services that daemonize. Services that daemonize themselves do not require the presence of any external service.
  • The supervisord-program container refers to the process supervisor that manages the services deployed by the Nix process management framework. It can also be used as a container for processes deployed by Disnix.

With the above infrastructure model, we can deploy any system that depends on the above container services, such as the trivial Disnix proxy example:

{ system, distribution, invDistribution, pkgs
, stateDir ? "/var"
, runtimeDir ? "${stateDir}/run"
, logDir ? "${stateDir}/log"
, cacheDir ? "${stateDir}/cache"
, tmpDir ? (if stateDir == "/var" then "/tmp" else "${stateDir}/tmp")
, forceDisableUserChange ? false
, processManager ? "supervisord"
, nix-processmgmt ? ../../../nix-processmgmt
}:

let
  customPkgs = import ../top-level/all-packages.nix {
    inherit system pkgs stateDir logDir runtimeDir tmpDir forceDisableUserChange processManager nix-processmgmt;
  };

  ids = if builtins.pathExists ./ids.nix then (import ./ids.nix).ids else {};

  processType = import "${nix-processmgmt}/nixproc/derive-dysnomia-process-type.nix" {
    inherit processManager;
  };
in
rec {
  hello_world_server = rec {
    name = "hello_world_server";
    port = ids.ports.hello_world_server or 0;
    pkg = customPkgs.hello_world_server { inherit port; };
    type = processType;
    requiresUniqueIdsFor = [ "ports" ];
  };

  hello_world_client = {
    name = "hello_world_client";
    pkg = customPkgs.hello_world_client;
    dependsOn = {
      inherit hello_world_server;
    };
    type = "package";
  };
}

The services model shown above (services.nix) captures two services:

  • The hello_world_server service is a simple service that listens on a TCP port for a "hello" message and responds with a "Hello world!" message.
  • The hello_world_client service is a package providing a client executable that automatically connects to the hello_world_server.

With the following distribution model (distribution.nix), we can map all the services to our deployment machine (that runs the Disnix service managed by the Nix process management framework):

{infrastructure}:

{
  hello_world_client = [ infrastructure.test1 ];
  hello_world_server = [ infrastructure.test1 ];
}

and deploy the system by running the following command:

$ disnix-env -s services-without-proxy.nix \
  -i infrastructure.nix \
  -d distribution.nix \
  --extra-params '{ processManager = "supervisord"; }'

The last parameter: --extra-params configures the services model (that indirectly invokes the createManagedProcess abstraction function from the Nix process management framework) in such a way that supervisord configuration files are generated.

(As a sidenote: without the --extra-params parameter, the process instances will be built for the disnix process manager generating configuration files that can be deployed to the process container, expecting programs to daemonize on their own and leave a PID file behind with the daemon's process ID. Although this approach is convenient for experiments, because no external service is required, it is not as reliable as managing supervised processes).

The result of the above deployment operation is that the hello-world-service service is deployed as a service that is also managed by supervisord:

$ supervisorctl 
dbus-daemon                      RUNNING   pid 2374, uptime 0:09:39
disnix-service                   RUNNING   pid 2397, uptime 0:09:38
hello-world-server               RUNNING   pid 2574, uptime 0:00:06
sshd                             RUNNING   pid 2375, uptime 0:09:39

and we can use the hello-world-client executable on the target machine to connect to the service:

$ /nix/var/nix/profiles/disnix/default/bin/hello-world-client 
Trying 192.168.2.1...
Connected to 192.168.2.1.
Escape character is '^]'.
hello
Hello world!

Deploying container providers and exposing them


With Disnix, it is also possible to deploy systems that are composed of different kinds of components, such as web services and databases.

For example, the Java variant of the ridiculous Staff Tracker example consists of the following services:


The services in the diagram above have the following purpose:

  • The StaffTracker service is the front-end web application that shows an overview of staff members and their locations.
  • The StaffService service is web service with a SOAP interface that provides read and write access to the staff records. The staff records are stored in the staff database.
  • The RoomService service provides read access to the rooms records, that are stored in a separate rooms database.
  • The ZipcodeService service provides read access to zip codes, that are stored in a separate zipcodes database.
  • The GeolocationService infers the location of a staff member from its IP address using the GeoIP service.

To deploy the system shown above, we need a target machine that provides Apache Tomcat (for managing the web application front-end and web services) and MySQL (for managing the databases) as container provider services:

{ pkgs ? import <nixpkgs> { inherit system; }
, system ? builtins.currentSystem
, stateDir ? "/var"
, runtimeDir ? "${stateDir}/run"
, logDir ? "${stateDir}/log"
, spoolDir ? "${stateDir}/spool"
, cacheDir ? "${stateDir}/cache"
, tmpDir ? (if stateDir == "/var" then "/tmp" else "${stateDir}/tmp")
, forceDisableUserChange ? false
, processManager
}:

let
  ids = if builtins.pathExists ./ids-tomcat-mysql.nix then (import ./ids-tomcat-mysql.nix).ids else {};

  constructors = import ../../services-agnostic/constructors.nix {
    inherit pkgs stateDir runtimeDir logDir tmpDir cacheDir spoolDir forceDisableUserChange processManager ids;
  };

  containerProviderConstructors = import ../../service-containers-agnostic/constructors.nix {
    inherit pkgs stateDir runtimeDir logDir tmpDir cacheDir spoolDir forceDisableUserChange processManager ids;
  };
in
rec {
  sshd = {
    pkg = constructors.sshd {
      extraSSHDConfig = ''
        UsePAM yes
      '';
    };

    requiresUniqueIdsFor = [ "uids" "gids" ];
  };

  dbus-daemon = {
    pkg = constructors.dbus-daemon {
      services = [ disnix-service ];
    };

    requiresUniqueIdsFor = [ "uids" "gids" ];
  };

  tomcat = containerProviderConstructors.simpleAppservingTomcat {
    commonLibs = [ "${pkgs.mysql_jdbc}/share/java/mysql-connector-java.jar" ];
    webapps = [
      pkgs.tomcat9.webapps # Include the Tomcat example and management applications
    ];

    properties.requiresUniqueIdsFor = [ "uids" "gids" ];
  };

  mysql = containerProviderConstructors.mysql {
    properties.requiresUniqueIdsFor = [ "uids" "gids" ];
  };

  disnix-service = {
    pkg = constructors.disnix-service {
      inherit dbus-daemon;
      containerProviders = [ tomcat mysql ];
    };

    requiresUniqueIdsFor = [ "gids" ];
  };
}

The process model above is an extension of the previous processes model, adding two container provider services:

  • tomcat is the Apache Tomcat server. The constructor function: simpleAppServingTomcat composes a configuration for a supported process manager, such as supervisord.

    Moreover, it bundles a Dysnomia container configuration file, and a Dysnomia module: tomcat-webapplication that can be used to manage the life-cycles of Java web applications embedded in the servlet container.
  • mysql is the MySQL DBMS server. The constructor function also creates a process manager configuration file, and bundles a Dysnomia container configuration file and module that manages the life-cycles of databases.
  • The container services above are propagated as containerProviders to the disnix-service. This function parameter is used to update the search paths for container configuration and modules, so that services can be deployed to these containers by Disnix.

After deploying the above processes model, we should see the following infrastructure model after capturing it:

$ disnix-capture-infra infra-bootstrap.nix
{
  "test1" = {
    properties = {
      "hostname" = "192.168.2.1";
      "system" = "x86_64-linux";
    };
    containers = {
      echo = {
      };
      fileset = {
      };
      process = {
      };
      supervisord-program = {
        "supervisordTargetDir" = "/etc/supervisor/conf.d";
      };
      wrapper = {
      };
      tomcat-webapplication = {
        "tomcatPort" = "8080";
        "catalinaBaseDir" = "/var/tomcat";
      };
      mysql-database = {
        "mysqlPort" = "3306";
        "mysqlUsername" = "root";
        "mysqlPassword" = "";
        "mysqlSocket" = "/var/run/mysqld/mysqld.sock";
      };
    };
    "system" = "x86_64-linux";
  };
}

As may be observed, the tomcat-webapplication and mysql-database containers (with their relevant configuration properties) were added to the infrastructure model.

With the following command we can deploy the example system's services to the containers in the network:

$ disnix-env -s services.nix -i infrastructure.nix -d distribution.nix

resulting in a fully functional system:


Deploying multiple container provider instances


As explained in the introduction, a limitation of the NixOS module system is that it is only possible to construct one instance of a service on a machine.

Process instances in a processes model deployed by the Nix process management framework as well as services in a Disnix services model are instantiated from functions that make it possible to deploy multiple instances of the same service to the same machine, by making conflicting properties configurable.

The following processes model was modified from the previous example to deploy two MySQL servers and two Apache Tomcat servers to the same machine:

{ pkgs ? import <nixpkgs> { inherit system; }
, system ? builtins.currentSystem
, stateDir ? "/var"
, runtimeDir ? "${stateDir}/run"
, logDir ? "${stateDir}/log"
, spoolDir ? "${stateDir}/spool"
, cacheDir ? "${stateDir}/cache"
, tmpDir ? (if stateDir == "/var" then "/tmp" else "${stateDir}/tmp")
, forceDisableUserChange ? false
, processManager
}:

let
  ids = if builtins.pathExists ./ids-tomcat-mysql-multi-instance.nix then (import ./ids-tomcat-mysql-multi-instance.nix).ids else {};

  constructors = import ../../services-agnostic/constructors.nix {
    inherit pkgs stateDir runtimeDir logDir tmpDir cacheDir spoolDir forceDisableUserChange processManager ids;
  };

  containerProviderConstructors = import ../../service-containers-agnostic/constructors.nix {
    inherit pkgs stateDir runtimeDir logDir tmpDir cacheDir spoolDir forceDisableUserChange processManager ids;
  };
in
rec {
  sshd = {
    pkg = constructors.sshd {
      extraSSHDConfig = ''
        UsePAM yes
      '';
    };

    requiresUniqueIdsFor = [ "uids" "gids" ];
  };

  dbus-daemon = {
    pkg = constructors.dbus-daemon {
      services = [ disnix-service ];
    };

    requiresUniqueIdsFor = [ "uids" "gids" ];
  };

  tomcat-primary = containerProviderConstructors.simpleAppservingTomcat {
    instanceSuffix = "-primary";
    httpPort = 8080;
    httpsPort = 8443;
    serverPort = 8005;
    ajpPort = 8009;
    commonLibs = [ "${pkgs.mysql_jdbc}/share/java/mysql-connector-java.jar" ];
    webapps = [
      pkgs.tomcat9.webapps # Include the Tomcat example and management applications
    ];
    properties.requiresUniqueIdsFor = [ "uids" "gids" ];
  };

  tomcat-secondary = containerProviderConstructors.simpleAppservingTomcat {
    instanceSuffix = "-secondary";
    httpPort = 8081;
    httpsPort = 8444;
    serverPort = 8006;
    ajpPort = 8010;
    commonLibs = [ "${pkgs.mysql_jdbc}/share/java/mysql-connector-java.jar" ];
    webapps = [
      pkgs.tomcat9.webapps # Include the Tomcat example and management applications
    ];
    properties.requiresUniqueIdsFor = [ "uids" "gids" ];
  };

  mysql-primary = containerProviderConstructors.mysql {
    instanceSuffix = "-primary";
    port = 3306;
    properties.requiresUniqueIdsFor = [ "uids" "gids" ];
  };

  mysql-secondary = containerProviderConstructors.mysql {
    instanceSuffix = "-secondary";
    port = 3307;
    properties.requiresUniqueIdsFor = [ "uids" "gids" ];
  };

  disnix-service = {
    pkg = constructors.disnix-service {
      inherit dbus-daemon;
      containerProviders = [ tomcat-primary tomcat-secondary mysql-primary mysql-secondary ];
    };

    requiresUniqueIdsFor = [ "gids" ];
  };
}

In the above processes model, we made the following changes:

  • We have configured two Apache Tomcat instances: tomcat-primary and tomcat-secondary. Both instances can co-exist because they have been configured in such a way that they listen to unique TCP ports and have a unique instance name composed from the instanceSuffix.
  • We have configured two MySQL instances: mysql-primary and mysql-secondary. Similar to Apache Tomcat, they can both co-exist because they listen to unique TCP ports (e.g. 3306 and 3307) and have a unique instance name.
  • Both the primary and secondary instances of the above services are propagated to the disnix-service (with the containerProviders parameter) making it possible for a client to discover them.

After deploying the above processes model, we can run the following command to discover the machine's configuration:

$ disnix-capture-infra infra-bootstrap.nix
{
  "test1" = {
    properties = {
      "hostname" = "192.168.2.1";
      "system" = "x86_64-linux";
    };
    containers = {
      echo = {
      };
      fileset = {
      };
      process = {
      };
      supervisord-program = {
        "supervisordTargetDir" = "/etc/supervisor/conf.d";
      };
      wrapper = {
      };
      tomcat-webapplication-primary = {
        "tomcatPort" = "8080";
        "catalinaBaseDir" = "/var/tomcat-primary";
      };
      tomcat-webapplication-secondary = {
        "tomcatPort" = "8081";
        "catalinaBaseDir" = "/var/tomcat-secondary";
      };
      mysql-database-primary = {
        "mysqlPort" = "3306";
        "mysqlUsername" = "root";
        "mysqlPassword" = "";
        "mysqlSocket" = "/var/run/mysqld-primary/mysqld.sock";
      };
      mysql-database-secondary = {
        "mysqlPort" = "3307";
        "mysqlUsername" = "root";
        "mysqlPassword" = "";
        "mysqlSocket" = "/var/run/mysqld-secondary/mysqld.sock";
      };
    };
    "system" = "x86_64-linux";
  };
}

As may be observed, the infrastructure model contains two Apache Tomcat instances and two MySQL instances.

With the following distribution model (distribution.nix), we can divide each database and web application over the two container instances:

{infrastructure}:

{
  GeolocationService = {
    targets = [
      { target = infrastructure.test1;
        container = "tomcat-webapplication-primary";
      }
    ];
  };
  RoomService = {
    targets = [
      { target = infrastructure.test1;
        container = "tomcat-webapplication-secondary";
      }
    ];
  };
  StaffService = {
    targets = [
      { target = infrastructure.test1;
        container = "tomcat-webapplication-primary";
      }
    ];
  };
  StaffTracker = {
    targets = [
      { target = infrastructure.test1;
        container = "tomcat-webapplication-secondary";
      }
    ];
  };
  ZipcodeService = {
    targets = [
      { target = infrastructure.test1;
        container = "tomcat-webapplication-primary";
      }
    ];
  };
  rooms = {
    targets = [
      { target = infrastructure.test1;
        container = "mysql-database-primary";
      }
    ];
  };
  staff = {
    targets = [
      { target = infrastructure.test1;
        container = "mysql-database-secondary";
      }
    ];
  };
  zipcodes = {
    targets = [
      { target = infrastructure.test1;
        container = "mysql-database-primary";
      }
    ];
  };
}

Compared to the previous distribution model, the above model uses a more verbose notation for mapping services.

As explained in an earlier blog post, in deployments in which only a single container is deployed, services are automapped to the container that has the same name as the service's type. When multiple instances exist, we need to manually specify the container where the service needs to be deployed to.

After deploying the system with the following command:

$ disnix-env -s services.nix -i infrastructure.nix -d distribution.nix

we will get a running system with the following deployment architecture:


Using the Disnix web service for executing remote deployment operations


By default, Disnix uses SSH to communicate to target machines in the network. Disnix has a modular architecture and is also capable of communicating to target machines by other means, for example via NixOps, the backdoor client, D-Bus, and directly executing tasks on a local machine.

There is also an external package: DisnixWebService that remotely exposes all deployment operations from a web service with a SOAP API.

To use the DisnixWebService, we must deploy a Java servlet container (such as Apache Tomcat) with the DisnixWebService application, configured in such a way that it can connect to the disnix-service over the D-Bus system bus.

The following processes model is an extension of the non-multi containers Staff Tracker example, with an Apache Tomcat service that bundles the DisnixWebService:

{ pkgs ? import <nixpkgs> { inherit system; }
, system ? builtins.currentSystem
, stateDir ? "/var"
, runtimeDir ? "${stateDir}/run"
, logDir ? "${stateDir}/log"
, spoolDir ? "${stateDir}/spool"
, cacheDir ? "${stateDir}/cache"
, tmpDir ? (if stateDir == "/var" then "/tmp" else "${stateDir}/tmp")
, forceDisableUserChange ? false
, processManager
}:

let
  ids = if builtins.pathExists ./ids-tomcat-mysql.nix then (import ./ids-tomcat-mysql.nix).ids else {};

  constructors = import ../../services-agnostic/constructors.nix {
    inherit pkgs stateDir runtimeDir logDir tmpDir cacheDir spoolDir forceDisableUserChange processManager ids;
  };

  containerProviderConstructors = import ../../service-containers-agnostic/constructors.nix {
    inherit pkgs stateDir runtimeDir logDir tmpDir cacheDir spoolDir forceDisableUserChange processManager ids;
  };
in
rec {
  sshd = {
    pkg = constructors.sshd {
      extraSSHDConfig = ''
        UsePAM yes
      '';
    };

    requiresUniqueIdsFor = [ "uids" "gids" ];
  };

  dbus-daemon = {
    pkg = constructors.dbus-daemon {
      services = [ disnix-service ];
    };

    requiresUniqueIdsFor = [ "uids" "gids" ];
  };

  tomcat = containerProviderConstructors.disnixAppservingTomcat {
    commonLibs = [ "${pkgs.mysql_jdbc}/share/java/mysql-connector-java.jar" ];
    webapps = [
      pkgs.tomcat9.webapps # Include the Tomcat example and management applications
    ];
    enableAJP = true;
    inherit dbus-daemon;

    properties.requiresUniqueIdsFor = [ "uids" "gids" ];
  };

  apache = {
    pkg = constructors.basicAuthReverseProxyApache {
      dependency = tomcat;
      serverAdmin = "admin@localhost";
      targetProtocol = "ajp";
      portPropertyName = "ajpPort";

      authName = "DisnixWebService";
      authUserFile = pkgs.stdenv.mkDerivation {
        name = "htpasswd";
        buildInputs = [ pkgs.apacheHttpd ];
        buildCommand = ''
          htpasswd -cb ./htpasswd admin secret
          mv htpasswd $out
        '';
      };
      requireUser = "admin";
    };

    requiresUniqueIdsFor = [ "uids" "gids" ];
  };

  mysql = containerProviderConstructors.mysql {
    properties.requiresUniqueIdsFor = [ "uids" "gids" ];
  };

  disnix-service = {
    pkg = constructors.disnix-service {
      inherit dbus-daemon;
      containerProviders = [ tomcat mysql ];
      authorizedUsers = [ tomcat.name ];
      dysnomiaProperties = {
        targetEPR = "http://$(hostname)/DisnixWebService/services/DisnixWebService";
      };
    };

    requiresUniqueIdsFor = [ "gids" ];
  };
}

The above processes model contains the following changes:

  • The Apache Tomcat process instance is constructed with the containerProviderConstructors.disnixAppservingTomcat constructor function automatically deploying the DisnixWebService and providing the required configuration settings so that it can communicate with the disnix-service over the D-Bus system bus.

    Because the DisnixWebService requires the presence of the D-Bus system daemon, it is configured as a dependency for Apache Tomcat ensuring that it is started before Apache Tomcat.
  • Connecting to the Apache Tomcat server including the DisnixWebService requires no authentication. To secure the web applications and the DisnixWebService, I have configured an apache reverse proxy that forwards connections to Apache Tomcat using the AJP protocol.

    Moreover, the reverse proxy protects incoming requests by using HTTP basic authentication requiring a username and password.

We can use the following bootstrap infrastructure model to discover the machine's configuration:

{
  test1.properties.targetEPR = "http://192.168.2.1/DisnixWebService/services/DisnixWebService";
}

The difference between this bootstrap infrastructure model and the previous is that it uses a different connection property (targetEPR) that refers to the URL of the DisnixWebService.

By default, Disnix uses the disnix-ssh-client to communicate to target machines. To use a different client, we must set the following environment variables:

$ export DISNIX_CLIENT_INTERFACE=disnix-soap-client
$ export DISNIX_TARGET_PROPERTY=targetEPR

The above environment variables instruct Disnix to use the disnix-soap-client executable and the targetEPR property from the infrastructure model as a connection string.

To authenticate ourselves, we must set the following environment variables with a username and password:

$ export DISNIX_SOAP_CLIENT_USERNAME=admin
$ export DISNIX_SOAP_CLIENT_PASSWORD=secret

The following command makes it possible to discover the machine's configuration using the disnix-soap-client and DisnixWebService:

$ disnix-capture-infra infra-bootstrap.nix
{
  "test1" = {
    properties = {
      "hostname" = "192.168.2.1";
      "system" = "x86_64-linux";
      "targetEPR" = "http://192.168.2.1/DisnixWebService/services/DisnixWebService";
    };
    containers = {
      echo = {
      };
      fileset = {
      };
      process = {
      };
      supervisord-program = {
        "supervisordTargetDir" = "/etc/supervisor/conf.d";
      };
      wrapper = {
      };
      tomcat-webapplication = {
        "tomcatPort" = "8080";
        "catalinaBaseDir" = "/var/tomcat";
        "ajpPort" = "8009";
      };
      mysql-database = {
        "mysqlPort" = "3306";
        "mysqlUsername" = "root";
        "mysqlPassword" = "";
        "mysqlSocket" = "/var/run/mysqld/mysqld.sock";
      };
    };
    "system" = "x86_64-linux";
  }
  ;
}

After capturing the full infrastructure model, we can deploy the system with disnix-env if desired, using the disnix-soap-client to carry out all necessary remote deployment operations.

Miscellaneous: using Docker containers as light-weight virtual machines


As explained earlier in this blog post, the Nix process management framework is only a partial infrastructure deployment solution -- you still need to somehow obtain physical or virtual machines with a software distribution running the Nix package manager.

In a blog post written some time ago, I have explained that Docker containers are not virtual machines or even light-weight virtual machines.

In my previous blog post, I have shown that we can also deploy mutable Docker multi-process containers in which process instances can be upgraded without stopping the container.

The deployment workflow for upgrading mutable containers, is very machine-like -- NixOS has a similar workflow that consists of updating the machine configuration (/etc/nixos/configuration.nix) and running a single command-line instruction to upgrade machine (nixos-rebuild switch).

We can actually start using containers as VMs by adding another ingredient in the mix -- we can also assign static IP addresses to Docker containers.

With the following Nix expression, we can create a Docker image for a mutable container, using any of the processes models shown previously as the "machine's configuration":

let
  pkgs = import <nixpkgs> {};

  createMutableMultiProcessImage = import ../nix-processmgmt/nixproc/create-image-from-steps/create-mutable-multi-process-image-universal.nix {
    inherit pkgs;
  };
in
createMutableMultiProcessImage {
  name = "disnix";
  tag = "test";
  contents = [ pkgs.mc pkgs.disnix ];
  exprFile = ./processes.nix;
  interactive = true;
  manpages = true;
  processManager = "supervisord";
}

The exprFile in the above Nix expression refers to a previously shown processes model, and the processManager the desired process manager to use, such as supervisord.

With the following command, we can build the image with Nix and load it into Docker:

$ nix-build
$ docker load -i result

With the following command, we can create a network to which our containers (with IP addresses) should belong:

$ docker network create --subnet=192.168.2.0/8 disnixnetwork

The above command creates a subnet with a prefix: 192.168.2.0 and allocates an 8-bit block for host IP addresses.

We can create and start a Docker container named: containervm using our previously built image, and assign it an IP address:

$ docker run --network disnixnetwork --ip 192.168.2.1 \
  --name containervm disnix:test

By default, Disnix uses SSH to connect to remote machines. With the following commands we can create a public-private key pair and copy the public key to the container:

$ ssh-keygen -t ed25519 -f id_test -N ""

$ docker exec containervm mkdir -m0700 -p /root/.ssh
$ docker cp id_test.pub containervm:/root/.ssh/authorized_keys
$ docker exec containervm chmod 600 /root/.ssh/authorized_keys
$ docker exec containervm chown root:root /root/.ssh/authorized_keys

On the coordinator machine, that carries out the deployment, we must add the private key to the SSH agent and configure the disnix-ssh-client to connect to the disnix-service:

$ ssh-add id_test
$ export DISNIX_REMOTE_CLIENT=disnix-client

By executing all these steps, containervm can be (mostly) used as if it were a virtual machine, including connecting to it with an IP address over SSH.

Conclusion


In this blog post, I have described how the Nix process management framework can be used as a partial infrastructure deployment solution for Disnix. It can be used both for deploying the disnix-service (to facilitate multi-user installations) as well as deploying container providers: services that manage the life-cycles of services deployed by Disnix.

Moreover, the Nix process management framework makes it possible to do these deployments on all kinds of software distributions that can use the Nix package manager, including NixOS, conventional Linux distributions and other operating systems, such as macOS and FreeBSD.

If I had developed this solution a couple of years ago, it would probably have saved me many hours of preparation work for my first demo in my NixCon 2015 talk in which I wanted demonstrate that it is possible to deploy services to a heterogeneous network that consists of a NixOS, Ubuntu and Windows machine. Back then, I had to do all the infrastructure deployment tasks manually.

I also have to admit (but this statement is mostly based on my personal preferences, not facts), is that I find the functional style that the framework uses is IMO far more intuitive than the NixOS module system for certain service configuration aspects, especially for configuring container services and exposing them with Disnix and Dysnomia:

  • Because every process instance is constructed from a constructor function that makes all instance parameters explicit, you are guarded against common configuration errors such as undeclared dependencies.

    For example, the DisnixWebService-enabled Apache Tomcat service requires access to the dbus-service providing the system bus. Not having this service in the processes model, causes a missing function parameter error.
  • Function parameters in the processes model make it more clear that a process depends on another process and what that relationship may be. For example, with the containerProviders parameter it becomes IMO really clear that the disnix-service uses them as potential deployment targets for services deployed by Disnix.

    In comparison, the implementations of the Disnix and Dysnomia NixOS modules are far more complicated and monolithic -- the Dysnomia module has to figure for all potential container services deployed as part of a NixOS configuration, their properties, convert them to Dysnomia configuration files, and configure the systemd configuration for the disnix-service for proper activation ordering.

    The wants parameter (used for activation ordering) is just a list of strings, not knowing whether it contains valid references to services that have been deployed already.

Availability


The constructor functions for the services as well as the deployment examples described in this blog post can be found in the Nix process management services repository.

Future work


Slowly more and more of my personal use cases are getting supported by the Nix process management framework.

Moreover, the services repository is steadily growing. To ensure that all the services that I have packaged so far do not break, I really need to focus my work on a service test solution.

Wednesday, February 24, 2021

Deploying mutable multi-process Docker containers with the Nix process management framework (or running Hydra in a Docker container)

In a blog post written several months ago, I have shown that the Nix process management framework can also be used to conveniently construct multi-process Docker images.

Although Docker is primarily used for managing single root application process containers, multi-process containers can sometimes be useful to deploy systems that consist of multiple, tightly coupled, processes.

The Docker manual has a section that describes how to construct images for multi-process containers, but IMO the configuration process is a bit tedious and cumbersome.

To make this process more convenient, I have built a wrapper function: createMultiProcessImage around the dockerTools.buildImage function (provided by Nixpkgs) that does the following:

  • It constructs an image that runs a Linux and Docker compatible process manager as an entry point. Currently, it supports supervisord, sysvinit, disnix and s6-rc.
  • The Nix process management framework is used to build a configuration for a system that consists of multiple processes, that will be managed by any of the supported process managers.

Although the framework makes the construction of multi-process images convenient, a big drawback of multi-process Docker containers is upgrading them -- for example, for Debian-based containers you can imperatively upgrade packages by connecting to the container:

$ docker exec -it mycontainer /bin/bash

and upgrade the desired packages, such as file:

$ apt install file

The upgrade instruction above is not reproducible -- apt may install file version 5.38 today, and 5.39 tomorrow.

To cope with these kinds of side-effects, Docker works with images that snapshot the outcomes of all the installation steps. Constructing a container from the same image will always provide the same versions of all dependencies.

As a consequence, to perform a reproducible container upgrade, it is required to construct a new image, discard the container and reconstruct the container from the new image version, causing the system as a whole to be terminated, including the processes that have not changed.

For a while, I have been thinking about this limitation and developed a solution that makes it possible to upgrade multi-process containers without stopping and discarding them. The only exception is the process manager.

To make deployments reproducible, it combines the reproducibility properties of Docker and Nix.

In this blog post, I will describe how this solution works and how it can be used.

Creating a function for building mutable Docker images


As explained in an earlier blog post, that compares the deployment properties of Nix and Docker, both solutions support reproducible deployment, albeit for different application domains.

Moreover, their reproducibility properties are built around different concepts:

  • Docker containers are reproducible, because they are constructed from images that consist of immutable layers identified by hash codes derived from their contents.
  • Nix package builds are reproducible, because they are stored in isolation in a Nix store and made immutable (the files' permissions are set read-only). In the construction process of the packages, many side effects are mitigated.

    As a result, when the hash code prefix of a package (derived from all build inputs) is the same, then the build output is also (nearly) bit-identical, regardless of the machine on which the package was built.

By taking these reproducibilty properties into account, we can create a reproducible deployment process for upgradable containers by using a specific separation of responsibilities.

Deploying the base system


For the deployment of the base system that includes the process manager, we can stick ourselves to the traditional Docker deployment workflow based on images (the only unconventional aspect is that we use Nix to build a Docker image, instead of Dockerfiles).

The process manager that the image provides deploys its configuration from a dynamic configuration directory.

To support supervisord, we can invoke the following command as the container's entry point:

supervisord --nodaemon \
  --configuration /etc/supervisor/supervisord.conf \
  --logfile /var/log/supervisord.log \
  --pidfile /var/run/supervisord.pid

The above command starts the supervisord service (in foreground mode), using the supervisord.conf configuration file stored in /etc/supervisord.

The supervisord.conf configuration file has the following structure:

[supervisord]

[include]
files=conf.d/*

The above configuration automatically loads all program definitions stored in the conf.d directory. This directory is writable and initially empty. It can be populated with configuration files generated by the Nix process management framework.

For the other process managers that the framework supports (sysvinit, disnix and s6-rc), we follow a similar strategy -- we configure the process manager in such a way that the configuration is loaded from a source that can be dynamically updated.

Deploying process instances


Deployment of the process instances is not done in the construction of the image, but by the Nix process management framework and the Nix package manager running in the container.

To allow a processes model deployment to refer to packages in the Nixpkgs collection and install binary substitutes, we must configure a Nix channel, such as the unstable Nixpkgs channel:

$ nix-channel --add https://nixos.org/channels/nixpkgs-unstable
$ nix-channel --update

(As a sidenote: it is also possible to subscribe to a stable Nixpkgs channel or a specific Git revision of Nixpkgs).

The processes model (and relevant sub models, such as ids.nix that contains numeric ID assignments) are copied into the Docker image.

We can deploy the processes model for supervisord as follows:

$ nixproc-supervisord-switch

The above command will deploy the processes model in the NIXPROC_PROCESSES environment variable, which defaults to: /etc/nixproc/processes.nix:

  • First, it builds supervisord configuration files from the processes model (this step also includes deploying all required packages and service configuration files)
  • It creates symlinks for each configuration file belonging to a process instance in the writable conf.d directory
  • It instructs supervisord to reload the configuration so that only obsolete processes get deactivated and new services activated, causing unchanged processes to remain untouched.

(For the other process managers, we have equivalent tools: nixproc-sysvinit-switch, nixproc-disnix-switch and nixproc-s6-rc-switch).

Initial deployment of the system


Because only the process manager is deployed as part of the image (with an initially empty configuration), the system is not yet usable when we start a container.

To solve this problem, we must perform an initial deployment of the system on first startup.

I used my lessons learned from the chainloading techniques in s6 (in the previous blog post) and developed hacky generated bootstrap script (/bin/bootstrap) that serves as the container's entry point:

cat > /bin/bootstrap <<EOF
#! ${pkgs.stdenv.shell} -e

# Configure Nix channels
nix-channel --add ${channelURL}
nix-channel --update

# Deploy the processes model (in a child process)
nixproc-${input.processManager}-switch &

# Overwrite the bootstrap script, so that it simply just
# starts the process manager the next time we start the
# container
cat > /bin/bootstrap <<EOR
#! ${pkgs.stdenv.shell} -e
exec ${cmd}
EOR

# Chain load the actual process manager
exec ${cmd}
EOF
chmod 755 /bin/bootstrap

The generated bootstrap script does the following:

  • First, a Nix channel is configured and updated so that we can install packages from the Nixpkgs collection and obtain substitutes.
  • The next step is deploying the processes model by running the nixproc-*-switch tool for a supported process manager. This process is started in the background (as a child process) -- we can use this trick to force the managing bash shell to load our desired process supervisor as soon as possible.

    Ultimately, we want the process manager to become responsible for supervising any other process running in the container.
  • After the deployment process is started in the background, the bootstrap script is overridden by a bootstrap script that becomes our real entry point -- the process manager that we want to use, such as supervisord.

    Overriding the bootstrap script makes sure that the next time we start the container, it will start instantly without attempting to deploy the system again.
  • Finally, the bootstrap script "execs" into the real process manager, becoming the new PID 1 process. When the deployment of the system is done (the nixproc-*-switch process that still runs in the background), the process manager becomes responsible for reaping it.

With the above script, the workflow of deploying an upgradable/mutable multi-process container is the same as deploying an ordinary container from a Docker image -- the only (minor) difference is that the first time that we start the container, it may take some time before the services become available, because the multi-process system needs to be deployed by Nix and the Nix process management framework.

A simple usage scenario


Similar to my previous blog posts about the Nix process management framework, I will use the trivial web application system to demonstrate how the functionality of the framework can be used.

The web application system consists of one or more webapp processes (with an embedded HTTP server) that only return static HTML pages displaying their identities.

An Nginx reverse proxy forwards incoming requests to the appropriate webapp instance -- each webapp service can be reached by using its unique virtual host value.

To construct a mutable multi-process Docker image with Nix, we can write the following Nix expression (default.nix):

let
  pkgs = import <nixpkgs> {};

  nix-processmgmt = builtins.fetchGit {
    url = https://github.com/svanderburg/nix-processmgmt.git;
    ref = "master";
  };

  createMutableMultiProcessImage = import "${nix-processmgmt}/nixproc/create-image-from-steps/create-mutable-multi-process-image-universal.nix" {
    inherit pkgs;
  };
in
createMutableMultiProcessImage {
  name = "multiprocess";
  tag = "test";
  contents = [ pkgs.mc ];
  exprFile = ./processes.nix;
  idResourcesFile = ./idresources.nix;
  idsFile = ./ids.nix;
  processManager = "supervisord"; # sysvinit, disnix, s6-rc are also valid options
}

The above Nix expression invokes the createMutableMultiProcessImage function that constructs a Docker image that provides a base system with a process manager, and a bootstrap script that deploys the multi-process system:

  • The name, tag, and contents parameters specify the image name, tag and the packages that need to be included in the image.
  • The exprFile parameter refers to a processes model that captures the configurations of the process instances that need to be deployed.
  • The idResources parameter refers to an ID resources model that specifies from which resource pools unique IDs need to be selected.
  • The idsFile parameter refers to an IDs model that contains the unique ID assignments for each process instance. Unique IDs resemble TCP/UDP port assignments, user IDs (UIDs) and group IDs (GIDs).
  • We can use the processManager parameter to select the process manager we want to use. In the above example it is supervisord, but other options are also possible.

We can use the following processes model (processes.nix) to deploy a small version of our example system:

{ pkgs ? import <nixpkgs> { inherit system; }
, system ? builtins.currentSystem
, stateDir ? "/var"
, runtimeDir ? "${stateDir}/run"
, logDir ? "${stateDir}/log"
, cacheDir ? "${stateDir}/cache"
, tmpDir ? (if stateDir == "/var" then "/tmp" else "${stateDir}/tmp")
, forceDisableUserChange ? false
, processManager
}:

let
  nix-processmgmt = builtins.fetchGit {
    url = https://github.com/svanderburg/nix-processmgmt.git;
    ref = "master";
  };

  ids = if builtins.pathExists ./ids.nix then (import ./ids.nix).ids else {};

  sharedConstructors = import "${nix-processmgmt}/examples/services-agnostic/constructors/constructors.nix" {
    inherit pkgs stateDir runtimeDir logDir cacheDir tmpDir forceDisableUserChange processManager ids;
  };

  constructors = import "${nix-processmgmt}/examples/webapps-agnostic/constructors/constructors.nix" {
    inherit pkgs stateDir runtimeDir logDir tmpDir forceDisableUserChange processManager ids;
  };
in
rec {
  webapp = rec {
    port = ids.webappPorts.webapp or 0;
    dnsName = "webapp.local";

    pkg = constructors.webapp {
      inherit port;
    };

    requiresUniqueIdsFor = [ "webappPorts" "uids" "gids" ];
  };

  nginx = rec {
    port = ids.nginxPorts.nginx or 0;

    pkg = sharedConstructors.nginxReverseProxyHostBased {
      webapps = [ webapp ];
      inherit port;
    } {};

    requiresUniqueIdsFor = [ "nginxPorts" "uids" "gids" ];
  };
}

The above Nix expression configures two process instances, one webapp process that returns a static HTML page with its identity and an Nginx reverse proxy that forwards connections to it.

A notable difference between the expression shown above and the processes models of the same system shown in my previous blog posts, is that this expression does not contain any references to files on the local filesystem, with the exception of the ID assignments expression (ids.nix).

We obtain all required functionality from the Nix process management framework by invoking builtins.fetchGit. Eliminating local references is required to allow the processes model to be copied into the container and deployed from within the container.

We can build a Docker image as follows:

$ nix-build

load the image into Docker:

$ docker load -i result

and create and start a Docker container:

$ docker run -it --name webapps --network host multiprocess:test
unpacking channels...
warning: Nix search path entry '/nix/var/nix/profiles/per-user/root/channels' does not exist, ignoring
created 1 symlinks in user environment
2021-02-21 15:29:29,878 CRIT Supervisor is running as root.  Privileges were not dropped because no user is specified in the config file.  If you intend to run as root, you can set user=root in the config file to avoid this message.
2021-02-21 15:29:29,878 WARN No file matches via include "/etc/supervisor/conf.d/*"
2021-02-21 15:29:29,897 INFO RPC interface 'supervisor' initialized
2021-02-21 15:29:29,897 CRIT Server 'inet_http_server' running without any HTTP authentication checking
2021-02-21 15:29:29,898 INFO supervisord started with pid 1
these derivations will be built:
  /nix/store/011g52sj25k5k04zx9zdszdxfv6wy1dw-credentials.drv
  /nix/store/1i9g728k7lda0z3mn1d4bfw07v5gzkrv-credentials.drv
  /nix/store/fs8fwfhalmgxf8y1c47d0zzq4f89fz0g-nginx.conf.drv
  /nix/store/vxpm2m6444fcy9r2p06dmpw2zxlfw0v4-nginx-foregroundproxy.sh.drv
  /nix/store/4v3lxnpapf5f8297gdjz6kdra8g7k4sc-nginx.conf.drv
  /nix/store/mdldv8gwvcd5fkchncp90hmz3p9rcd99-builder.pl.drv
  /nix/store/r7qjyr8vr3kh1lydrnzx6nwh62spksx5-nginx.drv
  /nix/store/h69khss5dqvx4svsc39l363wilcf2jjm-webapp.drv
  /nix/store/kcqbrhkc5gva3r8r0fnqjcfhcw4w5il5-webapp.conf.drv
  /nix/store/xfc1zbr92pyisf8lw35qybbn0g4f46sc-webapp.drv
  /nix/store/fjx5kndv24pia1yi2b7b2bznamfm8q0k-supervisord.d.drv
these paths will be fetched (78.80 MiB download, 347.06 MiB unpacked):
...

As may be noticed by looking at the output, on first startup the Nix process management framework is invoked to deploy the system with Nix.

After the system has been deployed, we should be able to connect to the webapp process via the Nginx reverse proxy:

$ curl -H 'Host: webapp.local' http://localhost:8080
<!DOCTYPE html>
<html>
  <head>
    <title>Simple test webapp</title>
  </head>
  <body>
    Simple test webapp listening on port: 5000
  </body>
</html>

When it is desired to upgrade the system, we can change the system's configuration by connecting to the container instance:

$ docker exec -it webapps /bin/bash

In the container, we can edit the processes.nix configuration file:

$ mcedit /etc/nixproc/processes.nix

and make changes to the configuration of the system. For example, we can change the processes model to include a second webapp process:

{ pkgs ? import <nixpkgs> { inherit system; }
, system ? builtins.currentSystem
, stateDir ? "/var"
, runtimeDir ? "${stateDir}/run"
, logDir ? "${stateDir}/log"
, cacheDir ? "${stateDir}/cache"
, tmpDir ? (if stateDir == "/var" then "/tmp" else "${stateDir}/tmp")
, forceDisableUserChange ? false
, processManager
}:

let
  nix-processmgmt = builtins.fetchGit {
    url = https://github.com/svanderburg/nix-processmgmt.git;
    ref = "master";
  };

  ids = if builtins.pathExists ./ids.nix then (import ./ids.nix).ids else {};

  sharedConstructors = import "${nix-processmgmt}/examples/services-agnostic/constructors/constructors.nix" {
    inherit pkgs stateDir runtimeDir logDir cacheDir tmpDir forceDisableUserChange processManager ids;
  };

  constructors = import "${nix-processmgmt}/examples/webapps-agnostic/constructors/constructors.nix" {
    inherit pkgs stateDir runtimeDir logDir tmpDir forceDisableUserChange processManager ids;
  };
in
rec {
  webapp = rec {
    port = ids.webappPorts.webapp or 0;
    dnsName = "webapp.local";

    pkg = constructors.webapp {
      inherit port;
    };

    requiresUniqueIdsFor = [ "webappPorts" "uids" "gids" ];
  };

  webapp2 = rec {
    port = ids.webappPorts.webapp2 or 0;
    dnsName = "webapp2.local";

    pkg = constructors.webapp {
      inherit port;
      instanceSuffix = "2";
    };

    requiresUniqueIdsFor = [ "webappPorts" "uids" "gids" ];
  };

  nginx = rec {
    port = ids.nginxPorts.nginx or 0;

    pkg = sharedConstructors.nginxReverseProxyHostBased {
      webapps = [ webapp webapp2 ];
      inherit port;
    } {};

    requiresUniqueIdsFor = [ "nginxPorts" "uids" "gids" ];
  };
}

In the above process model model, a new process instance named: webapp2 was added that listens on a unique port that can be reached with the webapp2.local virtual host value.

By running the following command, the system in the container gets upgraded:

$ nixproc-supervisord-switch

resulting in two webapp process instances running in the container:

$ supervisorctl 
nginx                            RUNNING   pid 847, uptime 0:00:08
webapp                           RUNNING   pid 459, uptime 0:05:54
webapp2                          RUNNING   pid 846, uptime 0:00:08
supervisor>

The first instance: webapp was left untouched, because its configuration was not changed.

The second instance: webapp2 can be reached as follows:

$ curl -H 'Host: webapp2.local' http://localhost:8080
<!DOCTYPE html>
<html>
  <head>
    <title>Simple test webapp</title>
  </head>
  <body>
    Simple test webapp listening on port: 5001
  </body>
</html>

After upgrading the system, the new configuration should also get reactivated after a container restart.

A more interesting example: Hydra


As explained earlier, to create upgradable containers we require a fully functional Nix installation in a container. This observation made a think about a more interesting example than the trivial web application system.

A prominent example of a system that requires Nix and is composed out of multiple tightly integrated process is Hydra: the Nix-based continuous integration service.

To make it possible to deploy a minimal Hydra service in a container, I have packaged all its relevant components for the Nix process management framework.

The processes model looks as follows:

{ pkgs ? import <nixpkgs> { inherit system; }
, system ? builtins.currentSystem
, stateDir ? "/var"
, runtimeDir ? "${stateDir}/run"
, logDir ? "${stateDir}/log"
, cacheDir ? "${stateDir}/cache"
, tmpDir ? (if stateDir == "/var" then "/tmp" else "${stateDir}/tmp")
, forceDisableUserChange ? false
, processManager
}:

let
  nix-processmgmt = builtins.fetchGit {
    url = https://github.com/svanderburg/nix-processmgmt.git;
    ref = "master";
  };

  nix-processmgmt-services = builtins.fetchGit {
    url = https://github.com/svanderburg/nix-processmgmt-services.git;
    ref = "master";
  };

  constructors = import "${nix-processmgmt-services}/services-agnostic/constructors.nix" {
    inherit nix-processmgmt pkgs stateDir runtimeDir logDir tmpDir cacheDir forceDisableUserChange processManager;
  };

  instanceSuffix = "";
  hydraUser = hydraInstanceName;
  hydraInstanceName = "hydra${instanceSuffix}";
  hydraQueueRunnerUser = "hydra-queue-runner${instanceSuffix}";
  hydraServerUser = "hydra-www${instanceSuffix}";
in
rec {
  nix-daemon = {
    pkg = constructors.nix-daemon;
  };

  postgresql = rec {
    port = 5432;
    postgresqlUsername = "postgresql";
    postgresqlPassword = "postgresql";
    socketFile = "${runtimeDir}/postgresql/.s.PGSQL.${toString port}";

    pkg = constructors.simplePostgresql {
      inherit port;
      authentication = ''
        # TYPE  DATABASE   USER   ADDRESS    METHOD
        local   hydra      all               ident map=hydra-users
      '';
      identMap = ''
        # MAPNAME       SYSTEM-USERNAME          PG-USERNAME
        hydra-users     ${hydraUser}             ${hydraUser}
        hydra-users     ${hydraQueueRunnerUser}  ${hydraUser}
        hydra-users     ${hydraServerUser}       ${hydraUser}
        hydra-users     root                     ${hydraUser}
        # The postgres user is used to create the pg_trgm extension for the hydra database
        hydra-users     postgresql               postgresql
      '';
    };
  };

  hydra-server = rec {
    port = 3000;
    hydraDatabase = hydraInstanceName;
    hydraGroup = hydraInstanceName;
    baseDir = "${stateDir}/lib/${hydraInstanceName}";
    inherit hydraUser instanceSuffix;

    pkg = constructors.hydra-server {
      postgresqlDBMS = postgresql;
      user = hydraServerUser;
      inherit nix-daemon port instanceSuffix hydraInstanceName hydraDatabase hydraUser hydraGroup baseDir;
    };
  };

  hydra-evaluator = {
    pkg = constructors.hydra-evaluator {
      inherit nix-daemon hydra-server;
    };
  };

  hydra-queue-runner = {
    pkg = constructors.hydra-queue-runner {
      inherit nix-daemon hydra-server;
      user = hydraQueueRunnerUser;
    };
  };

  apache = {
    pkg = constructors.reverseProxyApache {
      dependency = hydra-server;
      serverAdmin = "admin@localhost";
    };
  };
}

In the above processes model, each process instance represents a component of a Hydra installation:

  • The nix-daemon process is a service that comes with Nix package manager to facilitate multi-user package installations. The nix-daemon carries out builds on behalf of a user.

    Hydra requires it to perform builds as an unprivileged Hydra user and uses the Nix protocol to more efficiently orchestrate large builds.
  • Hydra uses a PostgreSQL database backend to store data about projects and builds.

    The postgresql process refers to the PostgreSQL database management system (DBMS) that is configured in such a way that the Hydra components are authorized to manage and modify the Hydra database.
  • hydra-server is the front-end of the Hydra service that provides a web user interface. The initialization procedure of this service is responsible for initializing the Hydra database.
  • The hydra-evaluator regularly updates the repository checkouts and evaluates the Nix expressions to decide which packages need to be built.
  • The hydra-queue-runner builds all jobs that were evaluated by the hydra-evaluator.
  • The apache server is used as a reverse proxy server forwarding requests to the hydra-server.

With the following commands, we can build the image, load it into Docker, and deploy a container that runs Hydra:

$ nix-build hydra-image.nix
$ docker load -i result
$ docker run -it --name hydra-test --network host hydra:test

After deploying the system, we can connect to the container:

$ docker exec -it hydra-test /bin/bash

and observe that all processes are running and managed by supervisord:

$ supervisorctl
apache                           RUNNING   pid 1192, uptime 0:00:42
hydra-evaluator                  RUNNING   pid 1297, uptime 0:00:38
hydra-queue-runner               RUNNING   pid 1296, uptime 0:00:38
hydra-server                     RUNNING   pid 1188, uptime 0:00:42
nix-daemon                       RUNNING   pid 1186, uptime 0:00:42
postgresql                       RUNNING   pid 1187, uptime 0:00:42
supervisor>

With the following commands, we can create our initial admin user:

$ su - hydra
$ hydra-create-user sander --password secret --role admin
creating new user `sander'

We can connect to the Hydra front-end in a web browser by opening http://localhost (this works because the container uses host networking):


and configure a job set to a build a project, such as libprocreact:


Another nice bonus feature of having multiple process managers supported is that if we build Hydra's Nix process management configuration for Disnix, we can also visualize the deployment architecture of the system with disnix-visualize:


The above diagram displays the following properties:

  • The outer box indicates that we are deploying to a single machine: localhost
  • The inner box indicates that all components are managed as processes
  • The ovals correspond to process instances in the processes model and the arrows denote dependency relationships.

    For example, the apache reverse proxy has a dependency on hydra-server, meaning that the latter process instance should be deployed first, otherwise the reverse proxy is not able to forward requests to it.

Building a Nix-enabled container image


As explained in the previous section, mutable Docker images require a fully functional Nix package manager in the container.

Since this may also be an interesting sub use case, I have created a convenience function: createNixImage that can be used to build an image whose only purpose is to provide a working Nix installation:

let
  pkgs = import <nixpkgs> {};

  nix-processmgmt = builtins.fetchGit {
    url = https://github.com/svanderburg/nix-processmgmt.git;
    ref = "master";
  };

  createNixImage = import "${nix-processmgmt}/nixproc/create-image-from-steps/create-nix-image.nix" {
    inherit pkgs;
  };
in
createNixImage {
  name = "foobar";
  tag = "test";
  contents = [ pkgs.mc ];
}

The above Nix expression builds a Docker image with a working Nix setup and a custom package: the Midnight Commander.

Conclusions


In this blog post, I have described a new function in the Nix process management framework: createMutableMultiProcessImage that creates reproducible mutable multi-process container images, by combining the reproducibility properties of Docker and Nix. With the exception of the process manager, process instances in a container can be upgraded without bringing the entire container down.

With this new functionality, the deployment workflow of a multi-process container configuration has become very similar to how physical and virtual machines are managed with NixOS -- you can edit a declarative specification of a system and run a single command-line instruction to deploy the new configuration.

Moreover, this new functionality allows us to deploy a complex, tightly coupled multi-process system, such as Hydra: the Nix-based continuous integration service. In the Hydra example case, we are using Nix for three deployment aspects: constructing the Docker image, deploying the multi-process system configuration and building the projects that are configured in Hydra.

A big drawback of mutable multi-process images is that there is no sharing possible between multiple multi-process containers. Since the images are not built from common layers, the Nix store is private to each container and all packages are deployed in the writable custom layer, this may lead to substantial disk and RAM overhead per container instance.

Deploying the processes model to a container instance can probably be made more convenient by using Nix flakes -- a new Nix feature that is still experimental. With flakes we can easily deploy an arbitrary number of Nix expressions to a container and pin the deployment to a specific version of Nixpkgs.

Another interesting observation is the word: mutable. I am not completely sure if it is appropriate -- both the layers of a Docker image, as well as the Nix store paths are immutable and never change after they have been built. For both solutions, immutability is an important ingredient in making sure that a deployment is reproducible.

I have decided to still call these deployments mutable, because I am looking at the problem from a Docker perspective -- the writable layer of the container (that is mounted on top of the immutable layers of an image) is modified each time that we upgrade a system.

Future work


Although I am quite happy with the ability to create mutable multi-process containers, there is still quite a bit of work that needs to be done to make the Nix process management framework more usable.

Most importantly, trying to deploy Hydra revealed all kinds of regressions in the framework. To cope with all these breaking changes, a structured testing approach is required. Currently, such an approach is completely absent.

I could also (in theory) automate the still missing parts of Hydra. For example, I have not automated the process that updates the garbage collector roots, which needs to run in a timely manner. To solve this, I need to use a cron service or systemd timer units, which is beyond the scope of my experiment.

Availability


The createMutableMultiProcessImage function is part of the experimental Nix process management framework GitHub repository that is still under heavy development.

Because the amount of services that can be deployed with the framework has grown considerably, I have moved all non-essential services (not required for testing) into a separate repository. The Hydra constructor functions can be found in this repository as well.