(OTS) Mempool GC Does Not Correctly Filter by Timestamp

[l-monninger]

Summary

Mempool GC does not correctly filter out transactions by timestamp. Each entry key has a specific format: {priority}:{timestamp}:{seqnum}:{id}which means that the key passed to the upper_bound call here does not make sense since it is formatted as {timestamp_threshold}: here. Which, instead of filtering out transactions based on timestamp will end filtering out transactions with priority<=timestamp_threshold.

[l-monninger]

@mzabaluev I think this may have resulted from #749 and #628 being in-flight at the same time.

[l-monninger]

@khokho

[mzabaluev]

@mzabaluev I think this may have resulted from #749 and #628 being in-flight at the same time.

Yes, I suppose this is exactly what happened.

I see the following alternative ways to remedy this:

• Erase the mempool priority altogether.
• Don't rely on global range indexing to implement GC.
  • A brute-force pass through all entries is the simplest solution, but may be inefficient if the number of pooled transactions is large.
  • What we can try to do instead is the combination of a cursor pass enumerating priority buckets as the outer loop, and prefix range sweeps for timestamp < timestamp_threshold in each priority bucket.

[mzabaluev]

And another one:

• Redesign the database so that there is the column family with timestamp as the top-level key, and another one with the total ordering.