[driv3r]

Fixes #214. Supersedes #215.

Problem

ProgressReportFrequency and StateReportFrequency previously had no defaults or validation. With a callback URI configured, zero or negative intervals made time.After fire immediately, repeatedly serializing and posting progress/state. The original copydb smoke observed 1,053 progress POSTs in 1.5 seconds when the frequency was omitted.

Policy

For each reporting frequency:

Configuration	Result
Omitted or 0	Use the 60000 ms default
Nonzero value below 100 ms, callback URI set	Return a validation error
100 ms or greater	Preserve the configured value
Callback URI empty	Reporting disabled; the minimum is not enforced

Zero is an unset/default sentinel, not a way to disable reporting. This follows the convention used by other optional numeric configuration fields. Explicit invalid nonzero values are rejected rather than silently clamped.

Library-path protection

The policy lives in one private Config.validateReportFrequencies helper, called by both Config.ValidateConfig and Ferry.Initialize.

Initialize invokes it after establishing the default error handler but before opening database connections. This protects the normal library Initialize → Start → Run lifecycle even when callers do not invoke Config.ValidateConfig, without duplicating policy in the reporting loops or revalidating unrelated settings inside Run.

Review updates

• Share the reporting checks with library initialization.
• Explicitly document zero/default semantics and URI-based disabling.
• Move the changelog entry to Changed because rejecting previously accepted nonzero values is a compatibility change.
• Replace the earlier frequency tests with a table-driven matrix for both fields: -1, 0, 99, 100, 500, and an unset callback URI. Tests assert acceptance/rejection and normalized or preserved values, not exact error wording.
• Add initialization regressions proving zero intervals are defaulted and invalid intervals fail before source/target connections are opened while leaving the error handler available.

The library-path regressions failed before this correction: initialization retained zero frequencies and accepted 99 ms. They pass with the shared helper.

Verification

• go test -count=1 -p=1 . ./test/go ./copydb/test ./sharding/test: all four suites passed.
• Ruby test/integration/callbacks_test.rb: four tests, zero failures.
• A standalone live library migration deliberately omitted Config.ValidateConfig. Initialize set both intervals to 60000 ms; the run copied three rows and emitted two progress POSTs and no state POSTs instead of flooding either callback.
• The same smoke program confirmed that 99 ms is rejected during initialization, with neither database connection opened and the error handler initialized.

Updated the branch with current main and retained all independent changelog entries when resolving the merge conflict.

CI follow-up: interrupt/resume shutdown

The failed Go 8.0/GTID job stopped before testing because pulling the MySQL image returned HTTP 502. The two Ruby 8.4 jobs instead timed out in the interrupt/resume suite.

Two test callbacks paused the data iterator with an unbounded sleep. After the Go subprocess exited, WEBrick waited forever for those Ruby request handlers, leaving the test harness stuck in kill while joining its server thread. A focused real-process/HTTP probe reproduced that exact shutdown state.

The callbacks now wait for the subprocess thread to finish, so they release when the child exits. A queue barrier ensures both affected regressions actually enter the paused-copy path before sending TERM; no data/verification assertions were removed, and no retries or longer job timeouts were added.

Post-fix verification used isolated Percona 8.4.4 servers with the CI server flags:

Configuration	CI seed	Full Ruby suite
zerolog / file_position	1695	76 tests, 0 failures
logrus / gtid	1881	76 tests, 0 failures

Both were single-process bundle exec rake test runs. The two forced-pause regressions also passed individually in both modes. The standalone shutdown probe changed from child_alive=false, shutdown_finished=false to child_alive=false, shutdown_finished=true.

Temporary database-port overrides used for isolated verification were restored before committing.

[shivnagarajan]

Just a few comments. Nothing super critical, but just some minor adjustments suggested.

[shivnagarajan]

it is a bit strange that 0 is handled differently from values < minReportFrequencyMs (100)

[driv3r]

Kept 0 as “unset/use the 60-second default,” consistent with optional Go integer fields so it wouldn't break existing setups, but apply new default