Keep Retry intervals given as a one-shot iterator - #2552

Problem

Retry(interval=...) accepts any iterable, but Retry.__init__() checks the values for negatives by looping over interval and then builds the list with list(interval). A generator, map() or iter() is used up by the check, so Retry.intervals ends up as [] with no error.

With retry=Retry(max=3, interval=<generator>) passed to enqueue(), the first time the job raises, Job.get_retry_interval() indexes the empty list and raises IndexError inside the worker's failure handling. The worker logs "found an unhandled exception, quitting..." and exits, and the job stays started. When it is later picked up as abandoned, StartedJobRegistry.cleanup() calls job.retry() and hits the same IndexError, so the next worker quits during its startup maintenance as well.

A job that returns Retry(interval=<generator>) doesn't crash anything, but Retry.get_interval() returns 0 for an empty list, so it is requeued at once with no backoff.

Reproduction

Real Redis 8.10.2, rq master (aacfa5b) and rq 2.12.0 from PyPI, default Worker, nothing mocked:

retry = Retry(max=3, interval=(10 * 2**i for i in range(3)))  # 10, 20, 40
retry.intervals  # []
job = queue.enqueue(always_fails, retry=retry)
queue.enqueue(always_fails)  # another job behind it
Worker([queue], connection=redis).work(max_idle_time=5)
# IndexError: list index out of range
# Worker ...: found an unhandled exception, quitting...
# work() returns at once; the first job is still `started`, the second still `queued`.

After the first job's entry in the started registry expired (about 90 seconds here), a new Worker quit the same way from clean_registries(). With a list, Retry(max=3, interval=[10, 20, 40]), the job is scheduled for retry as expected.

Change

Retry.__init__() now builds the list first and checks that, which is what Repeat.__init__() already does. Lists, tuples and ranges behave as before, and negative values are still rejected.

Verification

• The new test_retry_interval_from_iterator fails on master and passes with the fix, 10 runs each way. It also checks that a negative value in an iterator is still rejected and that a failing job is scheduled for retry.
• A separate check of the list form through the same worker path passes before and after.
• tests/test_retry.py: 18 passed before, 19 after. Full suite: 768 passed, 24 skipped on master; 769 passed, 24 skipped with the fix.
• ruff check and ruff format --check on the changed files are clean.

Summary by CodeRabbit

• Bug Fixes
  • Retry intervals supplied as one-shot iterators are now retained and applied correctly, allowing failed jobs to be scheduled for retry as configured.
  • Negative retry intervals continue to be rejected.