Vlad Lazar comments

Results 161 comments of


                                            Vlad Lazar

fix(walredo): yield while applying neon records

> overall this is insufficient because we have no signal when the basebackup requester dies even a single tenant with longer apply time than retry basebackup time could oom the...

lots of time wasted on `count_deltas()`

Update: * WAL based solution merged and will be released this week (2024-04-02) * Need to check things have improved post release. Keeping the ticket open until then.

lots of time wasted on `count_deltas()`

Looks like https://github.com/neondatabase/neon/pull/7230 helped here. Generated another flamegraph this morning and it's not exhibiting the original issue: (ask me if you want the svg - can't add it here for...

pageserver: Too early layer flushing

I believe this was fixed by https://github.com/neondatabase/neon/pull/6661. @koivunej do you agree?

pageserver: Too early layer flushing

Yeah, I agree. The issue still stands. > How to fix it in my mind, compare first write to an open inmemory layer instead of last flush? Not sure I...

pageserver: Too early layer flushing

We discussed this and concluded that we need a background task to periodically do the check. Where "check" is: > compare to when was the first write to the in-memory...

controller: graceful behavior for restarting all services at the same time

https://github.com/neondatabase/neon/issues/7797 should help here in on the happy path

controller: graceful behavior for restarting all services at the same time

Could we deal with this by updating docs: restart storage controller first and storage later last.

CI Failure (Too many attempts to find a segment to delete) in `CloudStorageScrubberTest.test_scrubber`

`sev/low` since it's a test issue

Many people report scheduler panics recently

@Noah-Kennedy rustc version was 1.76.0 Adding to what John said above, we also hit [this assertion](https://github.com/tokio-rs/tokio/blob/deff2524c354d3d3038e1c3813032701946a5c68/tokio/src/runtime/task/state.rs#L195-L200) in the scheduler. Can provide the stack trace, but I doubt it would be...