You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
With set prefer_olap = 'on' we observe process crashes in running TPC-H benchmark queries (for instance Q2) already at scale factor 10 in parallel with more than 10 clients on a single coordinator. The time until occurance of a crash strongly reduces with the number of clients. With more than 200 we observe them already after a few seconds. (If useful, we can provide you directly with scripts to reproduce this issue.)
It seems that memory gets corrupted. During a crash, always the first element of the memory freelist points to a non-accessible region (here to 0x10):
With
set prefer_olap = 'on'
we observe process crashes in running TPC-H benchmark queries (for instance Q2) already at scale factor 10 in parallel with more than 10 clients on a single coordinator. The time until occurance of a crash strongly reduces with the number of clients. With more than 200 we observe them already after a few seconds. (If useful, we can provide you directly with scripts to reproduce this issue.)It seems that memory gets corrupted. During a crash, always the first element of the memory freelist points to a non-accessible region (here to 0x10):
This results in a SIGSEV in the memory allocation.
Stack trace:
The database itself throws error messages like this:
The text was updated successfully, but these errors were encountered: