Reference ticket for the buster upgrade: T278137: Migrate eventlog1002 to buster
We currently run our legacy eventlogging on a single VM:
- eventlog1003.eqiad.wmnet
It runs the following eventlogging-processor services:
btullis@eventlog1003:~$ pstree -aT eventlogging python3 /srv/deployment/eventlogging/analytics/bin/eventlogging-processor @/etc/eventlogging.d/processors/client-side-07 python3 /srv/deployment/eventlogging/analytics/bin/eventlogging-processor @/etc/eventlogging.d/processors/client-side-01 python3 /srv/deployment/eventlogging/analytics/bin/eventlogging-processor @/etc/eventlogging.d/processors/client-side-09 python3 /srv/deployment/eventlogging/analytics/bin/eventlogging-processor @/etc/eventlogging.d/processors/client-side-05 python3 /srv/deployment/eventlogging/analytics/bin/eventlogging-processor @/etc/eventlogging.d/processors/client-side-04 python3 /srv/deployment/eventlogging/analytics/bin/eventlogging-processor @/etc/eventlogging.d/processors/client-side-11 python3 /srv/deployment/eventlogging/analytics/bin/eventlogging-processor @/etc/eventlogging.d/processors/client-side-10 python3 /srv/deployment/eventlogging/analytics/bin/eventlogging-processor @/etc/eventlogging.d/processors/client-side-08 python3 /srv/deployment/eventlogging/analytics/bin/eventlogging-processor @/etc/eventlogging.d/processors/client-side-02 python3 /srv/deployment/eventlogging/analytics/bin/eventlogging-processor @/etc/eventlogging.d/processors/client-side-03 python3 /srv/deployment/eventlogging/analytics/bin/eventlogging-processor @/etc/eventlogging.d/processors/client-side-00 python3 /srv/deployment/eventlogging/analytics/bin/eventlogging-processor @/etc/eventlogging.d/processors/client-side-06
However, the virtual machine is otherwise stateless.
All state is now stored in Kafka.
As per T278137, the recommended approach last time we need to upgrade was to create a parallel VM running the next O/S.
We then ran the two systems in parallel until we were confident enough that we could turn off the older version.
We may have to do some work on the eventlogging code to make sure that it works in the system python.
There is perhaps an argument here for skipping bullseye and moving straight to bookworm.
Tagging Event-Platform and Data-Engineering for visibility and in case they might be need to help update the code, but I believe that Data-Platform-SRE will provision the new VM and migrate the service when tested.