Open colega opened 1 month ago
This is an example I was giving someone on Slack:
Some series $S_1$, $S_2$ are sharded on ingester a-0
they're in-memory series.
Then we scale out ingesters and start a-1
. $S_2$ is now pushed to these replicas, while $S_1$ stays on a-0
(I'm obviously oversimplifying sharding).
Now a-0
has 2 in-memory series: $S_1$ and $S_2$ until TSDB is compacted and $S_2$ is garbage-collected, but it only has 1 owned series: $S_1$, because it doesn't own $S_2$ anymore.
If we sum(in-memory)
after scaling up, the result is 3: 2 from a-0
, 1 from a-1
, but customer is still pushing only 2 series.
Is your documentation request related to a feature? If so, which one?
In v2.12 we introduced an experimental feature called owned series tracking. There are however no docs explaining what are they, and why would someone want to use them instead of in-memory series.
Describe the solution that you’d like or the expected outcome
I would like to see docs explaining what owned series represent and what is the problem they're solving (I think the doc could start with the problem, and then introduce the concept of owned series).