Andrey Borodin
Andrey Borodin
Thanks for reporting this! I will think about how to make releases better... We can make brotli optional like lzo is now... but then we will have to ship lzo,br,lzo+br,...
Brotli can consume slightly more CPU, but we generally don't care as long as the restore is bottlenecked by network or FS. In most cases, brotli is the best codec...
Hi! AFAIK currently you can do PITR for each shard of MongoDB cluster. But if you restore each shard on some timestamp independently you may have some inconsistencies for transactions...
Maybe @perekalov can clarify why configserver ops are disabled?
We stripped binaries per @CyberDem0n request. Maybe we could also ship unstripped binary along with stripped? Ideally we should provide Lego-like executable constructor. User can request 1 compression method, 1...
Let's do daemon mode for wal-push? I'd propose even inheriting protocol from pgBackRest.
> @x4m > > Do we really need some protocol for Postgres at all? Couldn't it be a walg daemon that listens via a UNIX socket and the `wal-push` command...
I think daemonized version could employ parallelism. While usual WAL-G interface can do parallel upload only for sequential file names, not uploading same file many times.
I think that shipping many binaries is just fine. We already ship many versions, why not add one more? @matt-smiley can you plz adjust PR so that CI would build...
Look into postgres logs. Chances are you did not configure wal-push.