0.2.0

• Support for change stream replace events
• Removed the cluster time from the CEA cursor
• CSEvent now has a separate field for cluster time
• dripCEAResume: Tolerate resuming the "head" of the PCS
• dripCEAResume: Fix rejectIfOlderThan if the cursor is not on a nop
• Stop combining adjacent $match stages when inverting pipelines
• Persister: Avoid persisting all the changes in the cluster time, which used to lead to feedback loops.
• Support for Bun
• CC: Attach a comment to the MongoDB query
• Fix bug in rejectIfOlderThan logic

0.1.0

• Avoid using $or for tuple comparison. This sometimes confused MongoDB into using an in-memory sort, even if there is an index supporting the sort order. Instead, use multiple individual cursors and append the results.
• Reduce client traffic by only returning the noop event with the greatest cluster time, if it is more recent than all the other events, if any exist.
• Reduce database traffic by inverting the sync pipeline to distinguish updates from additions in select cases.
• Reduce database load by stripping away stages of the synced pipeline if possible.
• Reduce PCS size by not storing PCS event version in the PCS events themselves.
• Reduce PCS size by only storing the wall clock for noop events.
• Removed buggy support for $redact.
• Fixed buggy scoping for $replaceRoot and $replaceWith.
• Persister: Return an async generator. Allows it to be stopped gracefully.
• CEA: Support for refusing too old logical cursors. Useful for eventually removing old persisted change events without risking ongoing CEAs being affected.
• Added cleaner: Removes persisted change stream events older than a specified date.
• CEA: Removed some stages that cause MongoDB to always do an in-memory sort, even if there is an index supporting the sort order.
• Added support for processing pipelines. They run after the $sort stage but cannot filter out documents. Useful in cases where MongoDB resorts to an in-memory sort unnecessarily.
• CC: Return a lower bound for the cluster time at the beginning of CC.
• CC: Use causal consistency. This ensures CC never reads data older than the lower bound it returned.
• CEA: Take as parameter the lower bound for the cluster time at the beginning of CC instead of a wall clock. More efficient and does not assume that the client wall clock is accurate.
• Persister: Use bulk insertions and transactions to improve throughput.
• Close MongoDB cursors correctly.
• CC: Throw if there are no prior persisted events.

0.0.1

v0.0.1

Publish