server: allow configuring vmodule cluster-wide #89298
Assignees
Labels
A-cli-server
CLI commands that pertain to CockroachDB server processes
A-observability-inf
C-enhancement
Solution expected to add code/behavior + preserve backward-compat (pg compat issues are exception)
T-server-and-security
DB Server & Security
Comments
dt
added
C-enhancement
blathers-crl
bot
added
T-server-and-security
blathers-crl bot
pushed a commit
that referenced
this issue
Oct 18, 2022
Closes #89298. Release note: none. Epic: none.
This was referenced Oct 18, 2022
blathers-crl bot
pushed a commit
that referenced
this issue
Oct 18, 2022
Closes #89298. Release note: none. Epic: none.
craig bot
pushed a commit
that referenced
this issue
Oct 26, 2022
84888: distsql: remove queueing in the flow scheduler r=yuzefovich a=yuzefovich **flowinfra: de-duplicate cluster flow tests** Previously, there was a lot of duplication for two cluster flow tests (one for the single tenant setup and another for the multi tenant setup), and this commit cleans it up. Release note: None **distsql: remove queueing in the flow scheduler** This commit removes the queueing behavior in the flow scheduler (which is now renamed to "remote flow runner" to better reflect its purpose). This behavior was already disabled on 22.2 (with a cluster setting that could enable it back as an escape hatch) since we wanted to be conservative when removing it, but I don't foresee any problems with this, so it should be safe to remove it. We don't remove the "cancel dead flow coordinator" since it might still be useful in a mixed-version cluster, but more importantly the coordinator will be refactored to also cancel the running flows (not just the queued flows as it it used to). (This change will be in a separate commit.) This required removal some of the tests that relied on the queueing behavior, but I don't think we're losing much test coverage. Fixes: #34229. Release note (sql change): `sql.distsql.max_running_flows` cluster setting has been removed. This setting previously controlled the number of remote DistSQL flows that a single node would run at any time. Once that number was exceeded, the incoming flows would get queued until the number was reduced. This was used as a poor man's version of the admission control, but now that we have an actual admission control in place, we don't need that queueing behavior. 89324: roachtest: add admissioncontrol/index-overload r=andrewbaptist a=andrewbaptist This test sets up a 3 node cluster and measures the impact of creating an index while a controlled KV workload is running. The test measures two things * The baseline KV workload P99 latency * The impact on running index creation on the workload. The KV workload is designed to use about 20% of the CPU and IO resources of the system. Index creation is impactful by both reading a lot of data and writing a large index, however the primary impact is that it causes enough L0 inversion to make user traffic pause. Release note: None 89630: eval: clean up usage of UnwrapDatum r=yuzefovich a=yuzefovich This commit removes many calls to `eval.UnwrapDatum` where the first argument was `nil`. In such a case that function is equivalent to `tree.UnwrapDOidWrapper`, so this commit uses the latter wherever possible. Next, this commit adds an explicit `context.Context` argument to the signature to avoid the usage of the deprecated stored context from `eval.Context`. This required a little bit of plumbing around the index encoding methods. Release note: None Epic: None 89648: server: allow configuring vmodule via cluster setting r=dt a=dt Closes #89298. Release note: none. Epic: none. 90659: backfill: retain dropping columns when validating inverted index r=Xiang-Gu a=Xiang-Gu Previously, when validating a forward index, we made the first mutation(s) public with two filters: ignore constraints and retain dropping columns. But we forgot to include the retain-dropping-columns policy when validating *inverted* index. This will only manifest itself in rare cases involving dropping column and validating inverted indexes. An example to trigger this rare case is: ``` create table t (j int[], k int not null, inverted index (j)); alter table t alter primary key using columns (k); ``` Fixes #90306 Release note: None Co-authored-by: Yahor Yuzefovich <[email protected]> Co-authored-by: Andrew Baptist <[email protected]> Co-authored-by: David Taylor <[email protected]> Co-authored-by: Xiang Gu <[email protected]>
nicktrav
pushed a commit
to nicktrav/cockroach
that referenced
this issue
Nov 4, 2022
Closes cockroachdb#89298. Release note: none. Epic: none.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
setting the vmodule for some component is often very useful when debugging, but is sometimes difficult in large clusters, particularly when the interaction with them is through a load-balancer, as the builtin that can be used to set it at runtime only manipulates the in-memory state on the node that executes it.
We could augment the current setup by adding a cluster setting string such as
cluster.debug.vmodulethat can be set to the same string one would provide toset_vmodule. This cluster setting would have an on-change hook such that when it is updated, the value is applied, meaning all nodes update their vmodule when it is set or changed.Direct mutation of a single node's vmodule would still be possible via
set_vmoduleor the cli flag, overriding anything applied via the setting until the setting is next changed.Jira issue: CRDB-20195
The text was updated successfully, but these errors were encountered: