Remove old healthchecks (#57691)

.github/actions/setup-sentry/action.yml

@@ -185,4 +185,3 @@ runs:
         # This is necessary when other repositories (e.g. relay) want to take advantage of this workflow
         # without needing to fork it. The path needed is the one where tools are located
         cd "$WORKDIR"
-        python3 -u -S -m tools.devservices_healthcheck

tests/tools/test_devservices_healthcheck.py

@@ -1,110 +0,0 @@
-import subprocess
-import time
-from typing import Generator, List
-from unittest import mock
-
-import pytest
-
-from tools.devservices_healthcheck import HealthcheckError, check_health, run_with_retries
-
-
-@pytest.fixture(autouse=True)
-def no_sleep() -> Generator[None, None, None]:
-    with mock.patch.object(time, "sleep"):
-        yield
-
-
-@pytest.fixture
-def mock_subprocess_run() -> Generator[mock.Mock, None, None]:
-    with mock.patch.object(subprocess, "run", autospec=True) as mock_run:
-        yield mock_run
-
-
-def test_run_with_retries_fail() -> None:
-    with pytest.raises(HealthcheckError):
-        run_with_retries(
-            lambda: subprocess.run(("ls", "/tmp/this-does-not-exist"), check=True), 1, 10
-        )
-
-
-def test_run_with_retries_ok() -> None:
-    run_with_retries(lambda: subprocess.run("date", check=True), 1, 10)
-
-
-def test_postgres_not_running(mock_subprocess_run: mock.MagicMock) -> None:
-    mock_subprocess_run.return_value.stdout = ""
-    mock_subprocess_run.return_value.code = 0
-
-    with pytest.raises(HealthcheckError):
-        check_health(["postgres"])
-    assert mock_subprocess_run.call_count == 3
-
-
-def test_postgres_healthcheck_failing(mock_subprocess_run: mock.MagicMock) -> None:
-    running = mock.Mock()
-    running.stdout = "running\n"
-    running.code = 0
-
-    mock_subprocess_run.side_effect = [
-        running,
-        HealthcheckError("injected error"),
-        HealthcheckError("injected error"),
-        HealthcheckError("injected error"),
-    ]
-
-    with pytest.raises(HealthcheckError):
-        check_health(["postgres"])
-    assert mock_subprocess_run.call_count == 4
-
-
-def test_postgres_running(mock_subprocess_run: mock.MagicMock) -> None:
-    running = mock.Mock()
-    running.stdout = "running\n"
-    running.code = 0
-
-    healthcheck = mock.Mock()
-
-    mock_subprocess_run.side_effect = [
-        running,
-        healthcheck,
-    ]
-
-    check_health(["postgres"])
-    assert mock_subprocess_run.call_count == 2
-
-
-def test_kafka_running(mock_subprocess_run: mock.MagicMock) -> None:
-    running = mock.Mock()
-    running.stdout = "running\n"
-    running.code = 0
-
-    healthcheck = mock.Mock()
-
-    def run(
-        cmd_args: List[str], capture_output: bool = False, text: bool = False, check: bool = False
-    ) -> mock.Mock:
-        if cmd_args == (
-            "docker",
-            "container",
-            "inspect",
-            "-f",
-            "{{.State.Status}}",
-            "sentry_kafka",
-        ):
-            return running
-        elif cmd_args == (
-            "docker",
-            "exec",
-            "sentry_kafka",
-            "kafka-topics",
-            "--bootstrap-server",
-            "127.0.0.1:9092",
-            "--list",
-        ):
-            return healthcheck
-        raise AssertionError(f"unexpected command '{cmd_args}'")
-
-    mock_subprocess_run.side_effect = run
-
-    check_health(["kafka"])
-    assert mock_subprocess_run.call_count == 2

tools/devservices_healthcheck.py

@@ -1,163 +0,0 @@
-from __future__ import annotations
-
-import argparse
-import os
-import subprocess
-import time
-from collections.abc import Sequence
-from typing import Callable
-
-
-class HealthcheckError(Exception):
-    pass
-
-
-class HealthCheck:
-    def __init__(
-        self,
-        service_id: str,
-        container_name: str,
-        check_by_default: bool,
-        check: Callable[[], object] | None = None,
-        deps: list[str] | None = None,
-        retries: int = 3,
-        timeout_secs: int = 5,
-    ):
-        self.service_id = service_id
-        self.container_name = container_name
-        self.check_by_default = check_by_default
-        self.check = check
-        self.deps = deps or []
-        self.retries = retries
-        self.timeout_secs = timeout_secs
-
-    def check_container(self) -> None:
-        response = subprocess.run(
-            ("docker", "container", "inspect", "-f", "{{.State.Status}}", self.container_name),
-            capture_output=True,
-            text=True,
-        )
-        if response.stdout.strip() != "running":
-            raise HealthcheckError(f"Container '{self.container_name}' is not running.")
-
-
-def check_kafka():
-    subprocess.run(
-        (
-            "docker",
-            "exec",
-            "sentry_kafka",
-            "kafka-topics",
-            "--bootstrap-server",
-            "127.0.0.1:9092",
-            "--list",
-        ),
-        check=True,
-    )
-
-
-def check_postgres() -> None:
-    subprocess.run(
-        ("docker", "exec", "sentry_postgres", "pg_isready", "-U", "postgres"), check=True
-    )
-
-
-# Available health checks
-all_service_healthchecks = {
-    "postgres": HealthCheck(
-        "postgres",
-        "sentry_postgres",
-        True,
-        check_postgres,
-    ),
-    "kafka": HealthCheck(
-        "kafka",
-        "sentry_kafka",
-        os.getenv("NEED_KAFKA") == "true",
-        check_kafka,
-    ),
-}
-
-
-def run_with_retries(cmd: Callable[[], object], retries: int, timeout: int) -> None:
-    for retry in range(1, retries + 1):
-        try:
-            cmd()
-        except (HealthcheckError, subprocess.CalledProcessError) as e:
-            if retry == retries:
-                print(f"Command failed, no more retries: {e}")
-                raise HealthcheckError(f"Command failed: {e}")
-            else:
-                print(f"Command failed, retrying in {timeout}s (attempt {retry+1} of {retries})...")
-                time.sleep(timeout)
-        else:
-            return
-
-
-def get_services_to_check(id: str) -> list[str]:
-    checks = []
-    hc = all_service_healthchecks[id]
-    for dep in hc.deps:
-        dep_checks = get_services_to_check(dep)
-        for d in dep_checks:
-            checks.append(d)
-    checks.append(id)
-    return checks
-
-
-def check_health(service_ids: list[str]) -> None:
-    checks = [
-        check_id for service_id in service_ids for check_id in get_services_to_check(service_id)
-    ]
-
-    # dict.fromkeys is used to remove duplicates while maintaining order
-    unique_checks = list(dict.fromkeys(checks))
-    for name in unique_checks:
-        print(f"Checking service {name}")
-        hc = all_service_healthchecks[name]
-        print(f"Checking '{hc.container_name}' is running...")
-        ls = " ".join(unique_checks)
-        try:
-            run_with_retries(hc.check_container, hc.retries, hc.timeout_secs)
-        except HealthcheckError:
-            raise HealthcheckError(
-                f"Container '{hc.container_name}' is not running.\n"
-                f"    Start service: sentry devservices up {hc.service_id}\n"
-                f"    Restart all services: sentry devservices down {ls} && sentry devservices up {ls}"
-            )
-
-        if hc.check is not None:
-            print(f"Checking '{hc.container_name}' container health...")
-            try:
-                run_with_retries(hc.check, hc.retries, hc.timeout_secs)
-            except HealthcheckError:
-                raise HealthcheckError(
-                    f"Container '{hc.container_name}' does not appear to be healthy.\n"
-                    f"    Restart service: sentry devservices down {hc.service_id} && sentry devservices up {hc.service_id}\n"
-                    f"    Restart all services: sentry devservices down {ls} && sentry devservices up {ls}"
-                )
-
-
-def main(argv: Sequence[str] | None = None) -> None:
-    parser = argparse.ArgumentParser()
-    parser.add_argument(
-        "--service",
-        action="append",
-        choices=list(dict.fromkeys(all_service_healthchecks)),
-        help="The services you wish to check on. Defaults to all services.",
-    )
-
-    args = parser.parse_args(argv)
-
-    healthchecks = args.service
-    if healthchecks is None:
-        healthchecks = [k for k, v in all_service_healthchecks.items() if v.check_by_default]
-
-    try:
-        check_health(healthchecks)
-    except HealthcheckError as e:
-        raise SystemExit(e)
-
-
-if __name__ == "__main__":
-    main()