ADR-0032: Plugin Pool WaitGroup Drain + Context Cancellation

Status: Accepted
Date: 2026-06-12
Author: @johalputt

Context

The plugin goroutine pool was not draining cleanly on shutdown. Workers could be mid-execution when close(pluginQueue) was called, causing panics or lost work. Context cancellation was not propagated to plugin goroutines.

Decision

1. Add a sync.WaitGroup (workerPluginWg) that each plugin goroutine increments on start and decrements on exit.
2. Propagate the shutdown context to all plugin goroutines so they exit cleanly on cancellation.
3. On shutdown: close(pluginQueue) → workerPluginWg.Wait() to ensure all in-flight plugin jobs complete.
4. Each goroutine wraps its execution in recover() to isolate panics — a panicking plugin cannot crash the process.
5. pluginFailures uses atomic operations. pluginDisabled uses sync.Map.

Rationale

Without WaitGroup drain, a shutdown race where a plugin goroutine is mid-write to SQLite corrupts the write job state and leaves articles in processing limbo. The stuck-job reaper recovers these on next start, but it's operationally cleaner to drain first.

Consequences

• Positive: Clean shutdown; no stuck jobs from plugin mid-execution.
• Positive: Panic isolation — a bad plugin does not crash VayuPress.
• Negative: Shutdown may be delayed by up to one plugin job's processing time (~1s typical).