This would cut down on network I/O, binlog size, and replication overhead.
We already have a flags field in the table to support that. 0 means "legacy, assume gzipped if zlib present and gzinflate() works".
Description
Description
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Open | None | T255502 Goal: Save Timing median back under 1 second | |||
| Open | pmiazga | T302623 FY2022-2023: Improve Backend Pageview Timing | |||
| Open | None | T321966 Support igbinary with SqlBagOStuff |
Event Timeline
Comment Actions
SqlBagOStuff is powers the MainStash DB and the ParserCache DB. Motivation would be potentially reducing read and write times for those through less CPU work for encode/decode, and possibly less bandwidth.
Per @aaron the main motivation is actually not latency but infra cost. To decide on priority of this, I guess it's worth quantifying whether we are under pressure there and/or what kind of cost it would avoid or reduce. E.g. who benefits from this? (That is to say, reducing usage of a disk that can accomodate it already isn't worthwhile per-se.)