fix: reload log level on SIGHUP

[pollychen-lab]

Summary

Reload the active log level when pgstream receives SIGHUP instead of treating it as a shutdown signal.

Related Issue

Fixes #943

Changes

• Handles SIGHUP separately from shutdown signals.
• Reloads config and applies a changed logger level through zerolog's global level filter.
• Rejects invalid log levels and keeps the previous logger level in place.
• Avoids writing config reload notices to stdout during SIGHUP reloads.
• Adds focused tests for successful reload, invalid-level preservation, and quiet reload behavior.

Testing

• docker run --rm -v "$PWD":/src -w /src golang:1.26.4 gofmt -w cmd/root_cmd.go cmd/run_cmd.go cmd/snapshot_cmd.go cmd/root_cmd_test.go cmd/config/config.go internal/log/zerolog/zerolog.go pkg/log/zerolog/logger.go
• docker run --rm -v "$PWD":/src -w /src golang:1.26.4 go test ./cmd/... ./internal/log/... ./pkg/log/...
• docker run --rm -v "$PWD":/src -w /src golang:1.26.4 go test -race ./cmd
• git diff --check

Notes

Local host does not have Go installed, so validation was run in Docker with golang:1.26.4, matching the repository's Go version.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: fbe14e9e84

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Avoid writing config reload notices to stdout

When SIGHUP is used with a config file and the stdout target, this call goes through config.LoadFile, which prints using config file... with fmt.Printf to stdout; the stdout target writes NDJSON events to os.Stdout (pkg/wal/processor/stdout/wal_stdout_writer.go), so every reload injects a non-JSON line into the event stream. This is newly reachable at runtime because reload now calls config.Load() on each SIGHUP.

Useful? React with 👍 / 👎.

[pollychen-lab]

Good catch. I added a quiet config reload path and switched the SIGHUP handler to use that, so reloading a config file no longer emits the using config file... notice to stdout. I also added a regression test that captures stdout during reload and verifies it stays empty.

[chatgpt-codex-connector]

Guard logger mutation during SIGHUP reloads

When a SIGHUP arrives while run or snapshot is actively logging, the reload goroutine overwrites the shared *zerolog.Logger struct, while stdLogger and all WithFields loggers keep the same pointer and can read from it concurrently. That introduces a data race on the logger fields during normal logging; use a synchronized/atomic level mechanism or replace loggers without mutating the shared struct in place.

Useful? React with 👍 / 👎.

[pollychen-lab]

That makes sense. I changed the reload path so it no longer overwrites the shared *zerolog.Logger. The logger now keeps a stable trace-level instance and reload updates zerolog’s global level filter instead, which avoids mutating the logger struct while other goroutines may be logging. I also ran go test -race ./cmd successfully.