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
Hello guys,
we are using a very old version of Pulsar (2.9.X) (I am trying to convince management to upgrade it to newer, but its not an easy task, anyway) and we have services that use key-shared subscription that are configured to scale very fast on the workload. The load behaviour is very spiky with occasional spikes of x10, x15 times of the normal operating rates. Normally, I will implement a pre-heater and solve this issue, but it got me thinking...
How does and how fast key assignment of the key-shared subscriptions happen?
Is key assignment a client responsibility? For example if consumer is assigned keys A, B and C and has messages for A and B. Not a single message for C reached yet (not in client buffer, not after that, no prefetching or such), when new consumer joins the subscription will broker notice immediately that C is free for taking and immediately assigns C to it or does the original consumer that had this key assigned to first need to say something about it?
I am asking this because I notice certain pattern that whenever the spike arrives, no matter how aggressive is the scaling it always take affect few seconds after consumption already started so the faster the re-asignment happens the faster the scaling will be optimal
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Hello guys,
we are using a very old version of Pulsar (2.9.X) (I am trying to convince management to upgrade it to newer, but its not an easy task, anyway) and we have services that use key-shared subscription that are configured to scale very fast on the workload. The load behaviour is very spiky with occasional spikes of x10, x15 times of the normal operating rates. Normally, I will implement a pre-heater and solve this issue, but it got me thinking...
I am asking this because I notice certain pattern that whenever the spike arrives, no matter how aggressive is the scaling it always take affect few seconds after consumption already started so the faster the re-asignment happens the faster the scaling will be optimal
Beta Was this translation helpful? Give feedback.
All reactions