Opération #37

Titre Several http servers are slowed down
État terminé
Date de début 4 juin 2012 15:00
Date de fin 4 juin 2012 22:14
Serveurs concernés


4 juin 2012 15:17

An unexplained memory peak triggered a cache flush.

4 juin 2012 15:44

We suspect a SQL server was slowed down for a couple of minutes, which triggered the global memory usage increase and then, the cache flush on http servers.

There's no easy way to prevent this snowball effect from happening with our current architecture.

4 juin 2012 17:21

This problem happened again on http6. We're still unsure about the root cause.

4 juin 2012 18:11

We've probably found the culprit.