We recently installed libeatmydata on our test servers to improve the database performance (lots of resetting the database).
Libeatmydata works by patching libc's fsync function to return immediately instead of calling the kernel's syscall. It also does other similar changes to other functions.
Since then we have been experiencing some non-deterministic test failures. I can think of two explanations:
- Our tests or code are time-sensitive and
eatmydatamade it much faster. eatmydatasomehow changes the behavior of MySQL.
I believe (2) to be unlikely but I cannot find anything which suggests (1) in our code.
Do you have similar experience with mysql and libeatmydata?
Aucun commentaire:
Enregistrer un commentaire