StarRocks is a next-gen sub-second MPP database for full analytics scenarios, including multi-dimensional analytics, real-time analytics and ad-hoc query.
Problem:
Two problem which limit the download/upload speed:
If we use new authentication framework for S3, the client max connection is 25 by default and is not configurable.
backup/restore's download/upload task is limited that be can handle at most 3 task in a single backup/restore task.
Solutio...
After the materialized view is created, the FE reports an error when the data is synchronized regularly the next day, and then restarts,
Steps to reproduce the behavior (Required)
2023-06-05 14:59:10,553 WARN (stateChangeExecutor|68) [LakeTablet.getBackendIds():90] Failed to get backends by shard....
DynamicChunkBufferLimiter will alloc memory in pin method and may throw std::bad_alloc exception.
this method is invoked in ChunkSource::buffer_next_batch_chunks_blocking, if it throws std::bad_alloc exception ,BE will crash.
crash stack like this:
*** Aborted at 1685943281 (unix time) try "date...
if there is a memory allocation error in ChunkHelper::new_chunk_pooled, it will throw a std::bad_alloc exception, we should handle it.
otherwise, Be will crash, here is a stack as example
*** SIGABRT (@0x3e800007d3c) received by PID 32060 (TID 0x7fe53d8a2700) from PID 32060; stack trace: ***
@...
I tried to inject memory allocation error during query and found an crash issue.
Steps to reproduce the behavior (Required)
Expected behavior (Required)
Real behavior (Required)
StarRocks version (Required)
You can get the StarRocks version by executing SQL select current_version()
main e0a3851
Enhancement
mv: a left outer join b on a.c1 = b.c2
query: a inner join b on a.c1 = b.c2
query should be rewritten as:
select * from mv 1 where b.c2 is not null
or the compensation predicate is null-eliminated for right table