rptest: actual e2e throttling test #21379
Conversation
nvartolomei
force-pushed
the
nv/test_throttling
branch
from
5d48568 to
674e52c
Compare
|
skipped ducktape retry in https://buildkite.com/redpanda/redpanda/builds/51489#0190b09b-f50e-471c-9ed3-b0f6fbbf12e8: |
| return { | ||
| .disk_node_throughput_limit = tp + (tp / tput_overshoot_frac), | ||
| .disk_node_throughput_limit = tp, |
There was a problem hiding this comment.
Revert "cloud_storage: Increse disk tput limit a bit"
This reverts commit 0ae2ad2.
Needs explanation
There was a problem hiding this comment.
that was previous attempt to fix this test by increasing the disk limit a bit to make sure that the net limit is used to throttle downloads
There was a problem hiding this comment.
@nvartolomei please make sure that you put context into the commit messages. not having to visit github to see that context is a powerful thing.
nvartolomei
force-pushed
the
nv/test_throttling
branch
from
674e52c to
6f213fb
Compare
nvartolomei
force-pushed
the
nv/test_throttling
branch
from
6f213fb to
1975b7d
Compare
|
ducktape was retried in https://buildkite.com/redpanda/redpanda/builds/51581#0190bbe3-ac90-40b3-8b71-f89f335590ae ducktape was retried in https://buildkite.com/redpanda/redpanda/builds/51581#0190bbe3-ac92-4d20-a1ef-67f061a584a2 ducktape was retried in https://buildkite.com/redpanda/redpanda/builds/51581#0190bc6f-8f85-4dfd-9b3e-6e03db83ac61 |
rptest: actual e2e throttling test
Previous version of the test was setting an relatively low throughput
limit, then tried to consume data at full speed and monitor a metric to
see if tiered storage throttler applied any throttling.
This fails often because we actually throttle in the tiered storage
layer and we also apply disk throttling using seastar scheduling groups.
We can't measure if the later was applied and sometimes it provides
enough throttling that we never have to apply the tiered storage
throttling on the download path.
The new test is instead an e2e test. We consume a few times and measure
the average throughput. Then we throttle at half of that and measure a
few times again. The test succeeds if the second measurement was at half
of the first run.
Revert "cloud_storage: Increse disk tput limit a bit"
This reverts commit 0ae2ad2.
The original commit tried to relax the seastar i/o throttling in the
hope that the higher level throttler in tiered storage would be hit
instead which we can monitor (it does expose a metric). In practice this
wasn't enough.
The previous commit does rewrite the test so this hack is not required
anymore.
Closes #14139
Backports Required
Release Notes