Fix issue #689: Continuous snapshotting every second

[corylanou]

Summary

This PR fixes a critical bug where Litestream was continuously snapshotting every second instead of respecting the configured intervals, causing excessive S3 transactions (70,000+ overnight as reported).

Fixes #689

Root Cause

When snapshot.interval was not specified in the YAML config, the parser would set it to 0 (zero duration) which overwrote the 24h default. This caused time.Until() to return a negative duration, triggering immediate and continuous snapshots.

Solution

• Changed SnapshotConfig fields to use *time.Duration pointers to distinguish between "not specified" (nil) and "explicitly set to 0"
• Added comprehensive validation at config parse time to reject zero/negative intervals
• Preserved default values when YAML sections are empty or not specified
• Implemented sentinel errors for type-safe error checking

Changes Made

1. Configuration validation (cmd/litestream/main.go):

   • Added sentinel errors: ErrInvalidSnapshotInterval, ErrInvalidSnapshotRetention, ErrInvalidCompactionInterval, ErrInvalidSyncInterval
   • Added ConfigValidationError type with Unwrap() for errors.Is() support
   • Changed SnapshotConfig fields to use *time.Duration pointers
   • Added Validate() method to check all intervals > 0
   • Split ReadConfigFile into OpenConfigFile (io.ReadCloser) and ParseConfig (io.Reader)
   • Fixed default preservation when YAML sets fields to nil

2. Replica configuration (cmd/litestream/replicate.go):

   • Updated to handle pointer values for snapshot intervals
   • Only overrides defaults when explicitly set (non-nil)

3. Comprehensive tests (cmd/litestream/main_test.go):

   • Refactored complex table-driven tests to clear individual t.Run tests
   • Tests now use ParseConfig with strings.NewReader instead of temp files
   • Added validation for default values when sections are not specified
   • Tests for zero, negative, and unspecified intervals

Test Plan

✅ All existing tests pass
✅ New tests verify:

• Zero intervals are rejected with proper errors
• Negative intervals are rejected
• Default values (24h) are preserved when not specified
• Empty YAML sections don't overwrite defaults
• Multiple replicas with different sync intervals work correctly

Verification

Tested with a config similar to the reported issue:

dbs:
  - path: /tmp/test.db
    replicas:
      - url: s3://my-bucket/db
        sync-interval: 1800s

The sync-interval is properly set to 30m and snapshot interval defaults to 24h (not continuously snapshotting).

🤖 Generated with Claude Code