Time: 01/23/2020, from 5:45pm
Impact: ownCloud fast again
Dear users of GWDG ownCloud
we were able to attribute the failure of GWDG ownCloud in the morning of 23.01.2020. A
component in the processing of requests was too tightly designed.
This configuration has been revised, the service is available again.
In front of our database cluster we use a program that separates requests for read and
write access and forwards them to the appropriate nodes. By default, this program uses
only one CPU of the server. However, the machine had two CPUs, so even at 100% load, the
CPU usage could not increase above 50% and did not trigger our monitoring of the server.
The load reached this limit unnoticed which first led to occasional error messages but
in the course of the morning finally overwhelmed the service.
After the configuration was adjusted and more memory and CPU resources were made available
to the server, the service could be restarted.