See: https://github.com/NixOS/nixpkgs/blob/master/nixos/modules/services/admin/pgadmin.nix for inspiration

I found multiService useful for defining my local services, so ended up copying the code.

There should be comment in this file saying why we must maintain our own grafana package:

https://github.com/juspay/services-flake/blob/main/test/overlays/grafana.nix

Such a comment already exists on pgAdmin:

(Though here, this overlay should only apply on darwin).

https://discourse.nixos.org/t/prj-spec-direnv-lib/30294

Looks like PRJ_DATA_HOME could be used.

A just command that takes the service name as an argument and creates nix/<service-name>.nix and nix/<service-name>_test.nix with a hello-world service example.

• Maybe also add the files in the respective lists of nix/default.nix and test/flake.nix?

• Update README to add the service

• Comments specifying to take inspiration from either the nixos module or devenv module if either of them already implement the service.

Based on: https://github.com/cachix/devenv/blob/main/src/modules/services/clickhouse.nix

cf. "This also solves the problem of only being able to run one instance at a time." of cachix/devenv#75

Consider this scenario

passetto adds a postgresql service. But the user (nammayatri) may add their own postgresql service. Can the two DB coexist?

What we can do

Should we adjust the module API to accomodate for this? For eg., instead of:

services.postgresql.enable = true;

do we instead do:

services.postgresql."some-name".enable = true;

And consistently use this API for all services?

Concrete goal

This should work in nammayatri which has its own postgresql, as well as a postgres db used for passeto.

cf. #23 (comment)

Probably like this:

services.postgres.pg1 = {
  enable = true;
  initDumps = [
    ./dump1.sql
    ./dump2.sql
  ];
};

the following defaults:

are needed for the setup to function correctly and would rarely need to be overriden

a user however would often want to add more settings

in my case I'd like to add shared_preload_libraries = 'timescaledb'

we could add extraSettings which would be merged with settings or move current defaults to defaultSettings and leave settings empty, for the user to set

https://github.com/juspay/services-flake/actions/runs/7950883159/job/21703799091

I haven’t been able to reproduce it locally yet

We are on our way to making an initial release and doing formal announcement.

Checklist:

• Finish documentation (sufficiently)
• Create release tag, with empty CHANGELOG.md
  • For future PRs, setup guidelines to update the CHANGELOG
    • auto-generate CHANGELOG from commits. The commits will now follow specification from: https://www.conventionalcommits.org/en/v1.0.0/#summary, after: #121
• Blog post
• Announce.

After this release, we will record all changes in the CHANGELOG.md

Instead of initialScript have, say,

• initialScript.before
• initialScript.after

One runs before initialDatabases, and the other runs after.

...treefmt-check.drv logs from nixci:

treefmt-check> traversed 2 files
treefmt-check> matched 1 files to formatters
treefmt-check> left with 1 files after cache
treefmt-check> of whom 0 files were re-formatted
treefmt-check> all of this in 500ms

treefmt is only traversing the two files present in the dev sub dir.

Due to large number of requests to fetch the cache stored in magic-nix-cache, github responds with:

response body:

       GitHub API error: API error (429 Too Many Requests): StructuredApiError { message: "Request was blocked due to exceeding usage of resource 'Count' in namespace ''." }; retrying in 257 ms

triggering rebuild of several packages.

Possible solution

Can we use garnix cache instead?

Reference: https://github.com/cachix/devenv/blob/main/src/modules/services/elasticsearch.nix

https://github.com/juspay/passetto/pull/1/files#diff-d42ce6f4af2f6535fe936e696b7866ae0667acb573988e991dce1e76e03f84d1R34-R37

to reproduce

from the root of this repo:

1. check out https://github.com/rhinofi/services-flake/tree/postgres-superuser-issue (see diff from current main)
2. cd example
3. PC_DISABLE_TUI=true nix run

the output will be something like:

warning: Git tree '/code/services-flake' is dirty
[pg1-init	] The files belonging to this database system will be owned by user "adrian".
[pg1-init	] This user must also own the server process.
[pg1-init	] 
[pg1-init	] The database cluster will be initialized with locale "C".
[pg1-init	] The default text search configuration will be set to "english".
[pg1-init	] 
[pg1-init	] Data page checksums are disabled.
[pg1-init	] 
[pg1-init	] creating directory ./data/pg1 ... ok
[pg1-init	] creating subdirectories ... ok
[pg1-init	] selecting dynamic shared memory implementation ... posix
[pg1-init	] selecting default max_connections ... 100
[pg1-init	] selecting default shared_buffers ... 128MB
[pg1-init	] selecting default time zone ... Europe/London
[pg1-init	] creating configuration files ... ok
[pg1-init	] running bootstrap script ... ok
[pg1-init	] performing post-bootstrap initialization ... ok
[pg1-init	] syncing data to disk ... ok
[pg1-init	] 
[pg1-init	] 
[pg1-init	] Success. You can now start the database server using:
[pg1-init	] 
[pg1-init	]     /nix/store/ki3srrjjzqalvh0hd9lmqavp5v9wr9jp-postgresql-14.9/bin/pg_ctl -D ./data/pg1 -l logfile start
[pg1-init	] 
[pg1-init	] initdb: warning: enabling "trust" authentication for local connections
[pg1-init	] You can change this by editing pg_hba.conf or using the option -A, or
[pg1-init	] --auth-local and --auth-host, the next time you run initdb.
[pg1-init	] 
[pg1-init	] PostgreSQL initdb process complete.
[pg1-init	] 
[pg1-init	] Setting up postgresql.conf
[pg1-init	] 
[pg1-init	] PostgreSQL is setting up the initial database.
[pg1-init	] 
[pg1-init	] waiting for server to start....2024-02-21 16:15:28.951 GMT [668791] LOG:  starting PostgreSQL 14.9 on x86_64-pc-linux-gnu, compiled by gcc (GCC) 12.3.0, 64-bit
[pg1-init	] 2024-02-21 16:15:28.953 GMT [668791] LOG:  listening on Unix socket "/code/services-flake/example/data/pg1/pg-init-k2kvQC/.s.PGSQL.5432"
[pg1-init	] 2024-02-21 16:15:28.960 GMT [668792] LOG:  database system was shut down at 2024-02-21 16:15:28 GMT
[pg1-init	] 2024-02-21 16:15:28.963 GMT [668791] LOG:  database system is ready to accept connections
[pg1-init	]  done
[pg1-init	] server started
[pg1-init	] Checking presence of database: sample
[pg1-init	] psql: error: connection to server on socket "/code/services-flake/example/data/pg1/pg-init-k2kvQC/.s.PGSQL.5432" failed: FATAL:  role "adrian" does not exist
[pg1-init	] 2024-02-21 16:15:29.048 GMT [668803] FATAL:  role "adrian" does not exist
[pg1-init	] 0
[pg1-init	] Creating database: sample
[pg1-init	] 2024-02-21 16:15:29.056 GMT [668806] FATAL:  role "adrian" does not exist
[pg1-init	] psql: error: connection to server on socket "/code/services-flake/example/data/pg1/pg-init-k2kvQC/.s.PGSQL.5432" failed: FATAL:  role "adrian" does not exist

SQL dumps can have multi-line queries, for instance:

CREATE TABLE users (
  id SERIAL PRIMARY KEY,
  name VARCHAR(50) NOT NULL,
  email VARCHAR(50) NOT NULL UNIQUE
);

In such a case using postgres -E as here will error out because the command expects single line statements.

The solution to which can be postgres -j -E, but I think it is ideal to load the dumps with psql command instead -- as suggested in the postgres official doc.

Proposed solution

Context: The reason for using postgres is that we won't have the postgres server running while the init script is running.

We can run initialScript.before using postgres (as init script, before starting the server) and have initialScript.after run with psql as a process dependent on the postgres server.

Ideally, to keep things uniform, I would prefer running all init SQL commands via psql.

Required here: nammayatri/nammayatri#3718

Reference: https://github.com/cachix/devenv/blob/main/src/modules/services/nginx.nix

@adrian-gierakowski I think it would be nicer if there is one function that executes psql commands so that we don't accidentally add a psql command in the future that doesn't use psqlUserArg. That can be done in a different PR, merging this.

Originally posted by @shivaraj-bh in #106 (comment)

PC has a namespace column:

Which appears to come from the namespace yaml option,
https://github.com/search?q=repo%3AF1bonacc1%2Fprocess-compose%20namespace&type=code

I think it makes sense to put each process underservices.foo."name1" in its own namespace called "name1".

Support for this was added on Jun 10th: https://github.com/F1bonacc1/process-compose/releases/tag/v0.51.0 ... I wonder how namespaces work with other features, like depends_on.

It seems to happen while running default-test from example/flake.nix. This isn’t something that happens always but whenever it does, I will have to manually restart the workflow.

See an example workflow here: https://github.com/juspay/services-flake/actions/runs/7513430525/job/20455173179?pr=73

I can’t seem to reproduce this in my local atm.

It can be based on this: https://github.com/cachix/devenv/blob/main/src/modules/services/mysql.nix

For eg., run postgres (w/ chinhook db) + postgrest + pgAdmin in a flake.

Based on https://github.com/cachix/devenv/blob/main/src/modules/services/redis.nix

Goal: to be usable in the following,

1. nammayatri Backend
2. https://github.com/nammayatri/location-tracking-service (see PR #2 in that repo)

Test on (2) before merging.

To preempt issues like #12

• Depends on
  • #11
  • NixOS/nixpkgs#193336

We need this for nammayatri/location-tracking-service#34

The following services are missing docs:

• Apache Kafka
• [[clickhouse]]#
• Elasticsearch
• MySQL
• Nginx
• PostgreSQL
• Redis
• Redis Cluster
• Zookeeper
• [[grafana]]#
• [[prometheus]]#
• [[pgadmin]]#
• [[cassandra]]#

https://twitter.com/4ngerman/status/1765496606005268675

Related: #123

For eg., both pg1 and pg2 submodules can provide their own testScript

Do after #32

This type makes the user specify raw config (an error-prone process as well):

      services.clickhouse."clickhouse-db" = {
        extraConfig = ''
          http_port: 8123
        '';
      };

But ideally it should be using freeformType (pkgs.formats.yaml). See example here. Because, then the user can just write:

      services.clickhouse."clickhouse-db" = {
        extraConfig.http_port = 8123;
      };

We can't use Github Actions because that doesn't run nixosTests (unless using a self-hosted runner).

One way to host CI is to setup a Jenkins instance for github:juspay/* projects using https://github.com/juspay/jenkins-nix-ci and work on resolving juspay/jenkins-nix-ci#27 which we need to build the ./dev flake.

How to reproduce?

nix run github:shivaraj-bh/nammayatri/2bf9276b157f92f0b898852128b9632c95a97bb1#services

As can be seen in the screenshot above, the kafka process is still trying to connect to port 9092, even though the configuration sets port to 29092

Possible solution

I am not entirely sure of the cause but I will try to rebase the kafka service on this: https://github.com/NixOS/nixpkgs/blob/master/nixos/modules/services/misc/apache-kafka.nix

And additionally include tests for this scenario.

Both commitizen changelog and convco changelog can do this. Pick a suitable one, with appropriate arguments.

It would be good to see the changelog as we make conventional style commits right now, rather than get surprised during next release.

cd ./example
mkdir tmp
cd ./tmp
nix run

This creates the data diretory in ./example/tmp/data (instead of ./example/data as one would expect -- ie. in flake root). What is the best way to address this? In nammayatri we use flake-root + mission-control module to side-step the problem.

One possible solution here is to have services-flake require the use of flake-root module, and have the data directory use the flake root directory determined by the latter module.

I'd like to override default base dataDir for all services, without having to re-state service name in each override

I'd like to do this:

let
  baseDataDir = "./.data";
in {
  services.redis."redis-1" = {
    enable = true;
    inherit baseDataDir;
  };
  services.postgres."pg-1" = {
    enable = true;
    inherit baseDataDir;
  };
}

instead of:

let
  baseDataDir = "./.data";
in {
  services.redis."redis-1" = {
    enable = true;
    dataDir = "${baseDataDir}/redis-1";
  };
  services.postgres."pg-1" = {
    enable = true;
    dataDir = "${baseDataDir}/pg-1";
  };
}

while preserving the original data dir structure

see possible implementation here

The docs should link to https://nixos.asia/en/blog/replacing-docker-compose in an appropriate section.

The second commit in PR #127 has the commit message use same installer - yet the pre-commit check did not catch it, as it doesn't comfort to conventional commits spec.

https://community.flake.parts/services-flake/services

https://github.com/DeterminateSystems/magic-nix-cache

Unlike NixOS services, process-compose programs created by services-flake do not mutate outside of dataDir, so we gain very little from the isolation provided by VMs.

Let's switch to running these tests natively, which gets us macOS support (resolving #14), in addition to allowing us to use Github Actions for CI (#11).

Implementation idea

Use process-compose itself to run tests. process-compose running another process-compose, with readiness checks triggering the test process. We can use bash, or better nushell, to write the tests in. Nushell, because it can easily process the JSON from process-compose API.

This option will be useful to append extra depends_on configuration to the <name>-init process.

An example scenario is when a replica DB has to depend on another process that runs pg_basebackup.

For AI projects in Juspay, we are using a vector database called Weaviate. Need to add support for this in services-flake.

Nix package

• Expose options to configure the Weaviate server
• Enable startiing weaviate-server in a process-compose window upon nix run

To keep the tree autoformatted and check that in CI (as part of #11) as part of flake check.