Tomasz Grabiec
Tomasz Grabiec
Related thread: https://groups.google.com/forum/#!msg/scylladb-dev/r0bEtLO3bpQ/aQelmCdxCAAJ
@avikivity It's not mine.
Using https://github.com/RichardWarburton/honest-profiler/blob/master/src/main/java/com/insightfullogic/honest_profiler/ports/console/FlameGraphDumperApplication.java on the same file seems to work.
> What if a node in some region (or the entire region) is down? We may need to add a new node in our region, but now we can't, and...
Currently, remote coordinators assume they have accurate view on topology and replicate writes based on their view. If we bootstrap without updating them, we may end up losing writes due...
I don't understand the stall report: ``` | | | ++[3#1/1 100%] addr=0x128ac38 total=8862 count=6 avg=1477: | | | | table at ./replica/database.hh:776 | | | | (inlined by) seastar::lw_shared_ptr::make...
@bhalevy It's https://github.com/scylladb/scylladb/files/9598515/decoded-stalls-node-2.log from the first message. IMO, it's confusing to show total from the caller perspective if you group the tree from the callee perspective.
@soyacz Can you please attach the stall report for node-2 from the caller perspective?
I don't recall anything changing recently in this code path, so on the surface it doesn't look like regression.
@fgelcer It says `Use --direction=top-down to print callers first.`