ring/kv/dynamodb: reuse timers in watch loops to avoid per-poll allocations

[sandy2008]

What this PR does:

• Replaces time.After(c.pullerSyncTime) in DynamoDB watch loops with a reusable time.Timer.
• Applies this to both:
  • WatchKey()
  • WatchPrefix()
• Adds resetTimer() helper (stop + drain + reset) to safely reuse timers.
• Adds a focused benchmark to validate the change:
  • BenchmarkWatchLoopWaitWithTimeAfter
  • BenchmarkWatchLoopWaitWithReusableTimer

Performance benchmark on local machine shows:

• time.After: ~248 B/op, 3 allocs/op
• reusable timer: 0 B/op, 0 allocs/op

No behavior change intended, only allocation reduction in long-running watch loops.

Which issue(s) this PR fixes:
Fixes #

Checklist

• Tests updated
• Documentation added
• CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX]

[friedrichg]

Thanks for the PR

Adds a focused benchmark to validate the change:
BenchmarkWatchLoopWaitWithTimeAfter
BenchmarkWatchLoopWaitWithReusableTimer

I think you forgot to include these in the PR

[sandy2008]

Thanks for the PR

Adds a focused benchmark to validate the change:
BenchmarkWatchLoopWaitWithTimeAfter
BenchmarkWatchLoopWaitWithReusableTimer

I think you forgot to include these in the PR

@friedrichg Thx, added :)

[sandy2008]

Execution for the test:

• go test ./pkg/ring/kv/dynamodb

  • ok github.com/cortexproject/cortex/pkg/ring/kv/dynamodb (cached)

• go test ./pkg/ring/kv/dynamodb -run '^$' -bench 'BenchmarkWatchLoopWaitWith(TimeAfter|ReusableTimer)$' -benchmem -count=3

  • BenchmarkWatchLoopWaitWithTimeAfter: ~260–264 ns/op, 248 B/op, 3 allocs/op
  • BenchmarkWatchLoopWaitWithReusableTimer: ~248–259 ns/op, 0 B/op, 0 allocs/op
  • PASS

Result: reusable timer keeps latency similar and removes per-iteration allocations in the watch-loop wait path.

[sandy2008]

@friedrichg Sorry, reran modernize.