Add health check to docker compose

[willnode]

Added a health check so the service can heal from crashing by itself.

[ruslandoga]

👋 @willnode

How does it compare to #83? I'm afraid that setup would make the startup too slow (>30 seconds)

[willnode]

The healthcheck for the clickhouse is the recommended way of doing health check. Wait I'll benchmark it on my live server.

[ruslandoga]

The healthcheck for the clickhouse is the recommended way of doing health check.

Seems to be just a (possibly outdated) StackOverflow answer. Is there any official source recommending this exact approach? I'd think clickhouse client would be recommended now.

[willnode]

I'm afraid that setup would make the startup too slow (>30 seconds)

I can add these parameters (if you OK with fairly new docker compose requirement to 2.20 if i'm not mistaken)

    healthcheck:
      test: ...
      start_interval: 2s
      start_period: 1m

The startup from last down is like this now:

$ time docker compose up -d --wait
[+] Running 4/4
 ✔ Network public_html_default                  Created            0.1s
 ✔ Container public_html-plausible_db-1         Healthy            3.7s
 ✔ Container public_html-plausible_events_db-1  Healthy            3.7s
 ✔ Container public_html-plausible-1            Healthy           22.2s

real    0m22.428s
user    0m0.149s
sys     0m0.089s

If --wait omitted the startup would only need 3.7s

Is there any official source recommending this exact approach?

https://clickhouse.com/docs/en/interfaces/http

In health-check scripts use GET /ping request.

But both works and I see no difference in timing anyway.

[ruslandoga]

I can add these parameters

Yes, please do :)

[willnode]

Added

[ruslandoga]

👋 @willnode

I tried it on my instance for the last week, and all was great. Thank you!

[ruslandoga]

👋 @willnode

Sorry to bother you, but it seems like pg_isready is using root user, can you please verify? Should we edit the healthcheck to use $PG_USER or whatever it is? Or should we express the command as a list ["CMD", "pg_isready"], I remember there were some differences between that expression and a string which uses the shell, and possibly that's where the root user is coming from.

...
plausible_db-1  | 2024-08-02 10:35:25.439 UTC [575826] FATAL:  role "root" does not exist
plausible_db-1  | 2024-08-02 10:35:55.557 UTC [575835] FATAL:  role "root" does not exist
plausible_db-1  | 2024-08-02 10:36:25.681 UTC [575841] FATAL:  role "root" does not exist
plausible_db-1  | 2024-08-02 10:36:55.798 UTC [575849] FATAL:  role "root" does not exist
...

I'm reverting this PR for now to avoid confusing new users with these scary looking errors.

[ruslandoga]

Related to #146, but it's unlikely to be the cause.

[willnode]

Hello, sorry this change bothering you. I should really checking it from ground zero.

Here's my exact setup I currently use. I use docker rootless setup though I'm skeptical it has any difference. Maybe the version difference could be the cause?

  plausible_db:
    image: postgres:12-alpine
    restart: always
    volumes:
      - ./db-data:/var/lib/postgresql/data
    environment:
      - POSTGRES_PASSWORD=postgres
    healthcheck:
      test: pg_isready
      start_interval: 2s
      start_period: 1m

[BtbN]

This should be pretty easy to fix by just adding -U postgres to the pg_isready invocation.