Aljoscha Krettek comments

Results 80 comments of


                                            Aljoscha Krettek

[Epic] Surfacing Source/Sink Errors

We need to audit which source errors make it into the `Err` collection, and eventually into the persist shard. As of now, I think too many errors make it in...

[Epic] Surfacing Source/Sink Errors

This internal message (and thread) is important for things we need to thing about when we pick the work on this back up: https://materializeinc.slack.com/archives/C01CFKM1QRF/p1660726837927649. We need to figure out: -...

[Epic] Kafka Sinks (with EOS) for Platform

Nice! Some of them feel like they could still be issues. We just have to walk through each of them and try. 🙈

Alter Persist's row and error representation to avoid unused memory

cc @danhhz

performance: spikes in latency (>0.5-1s) for simple , non-concurrent SELECT queries

I asked @danhhz and/or @pH14 to quickly check if this is a persist problem. Based on that we can decide what to do.

performance: spikes in latency (>0.5-1s) for simple , non-concurrent SELECT queries

Hmm yes, I once referred to the retry pattern as "cas ping-ponging". :see_no_evil:

performance: spikes in latency (>0.5-1s) for simple , non-concurrent SELECT queries

I'm still wondering why that spike shows up in the queries: both advancing and downgrading happen in separate tasks right now, so they should not block coordinator work. Hypothesis: with...

performance: spikes in latency (>0.5-1s) for simple , non-concurrent SELECT queries

Just because I tried: when you run the tests with `SET transaction_isolation = 'serializable'` these slt tests fail.

performance: spikes in latency (>0.5-1s) for simple , non-concurrent SELECT queries

With things like ``` 00:00:00.000114 INSERT INTO tab0 VALUES(89,91,82) 00:00:00.036554 INSERT INTO tab0 VALUES(35,97,1) 00:00:00.015608 INSERT INTO tab0 VALUES(24,86,33) 00:00:00.017627 INSERT INTO tab1 VALUES(64,10,57) 00:00:00.017020 INSERT INTO tab1 VALUES(3,26,54) 00:00:00.033883...

performance: spikes in latency (>0.5-1s) for simple , non-concurrent SELECT queries

I mean it is correct. If you try for a bit more it will _eventually_ return the correct result. Side note: I don't know much about how it should behave...