[
https://issues.apache.org/jira/browse/IMPALA-13596?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17910444#comment-17910444
]
ASF subversion and git services commented on IMPALA-13596:
----------------------------------------------------------
Commit 21ef3e6ffe997c3dbb71e1d7a92844fb31387a92 in impala's branch
refs/heads/master from Andrew Sherman
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=21ef3e6ff ]
IMPALA-13638: Translate apostrophe to underscore in Prometheus metric names.
Impala has some metrics that reflect the state of the JVM. Some of these
metrics have names that are partly composed of the names of the
MemoryPoolMXBean objects in the Java virtual machine. In Jdk 8 these
are names like "Code Cache" and "PS Eden Space". In Jdk 11 these names
include apostrophe characters, for example "CodeHeap 'profiled
nmethods'". The derived metric names work OK for Impala in both the
webui and in json output. However the apostrophe character is illegal
in Prometheus metric names per
https://prometheus.io/docs/concepts/data_model/#metric-names-and-labels
and these metrics cannot be consumed by Prometheus. Fix this by adding
the apostrophe to the list of characters that are mapped to underscores
when we translate the metric names for Prometheus metrics.
TESTING:
Extended the test_prometheus_metrics test to parse all generated
Prometheus metrics. Ran the test with Jdk 11 where it failed without
the server fix
Change-Id: I557b123c075dff0b14ac527de08bc6177bd2a3f6
IMPALA-13596: first cut at tidied code
Reviewed-on: http://gerrit.cloudera.org:8080/22295
Reviewed-by: Impala Public Jenkins <[email protected]>
Tested-by: Impala Public Jenkins <[email protected]>
> Verify Admission Control configuration at startup
> -------------------------------------------------
>
> Key: IMPALA-13596
> URL: https://issues.apache.org/jira/browse/IMPALA-13596
> Project: IMPALA
> Issue Type: Bug
> Affects Versions: Impala 4.4.0
> Reporter: Andrew Sherman
> Assignee: Andrew Sherman
> Priority: Major
>
> The Admission Control configuration is split across the two files
> -fair_scheduler_allocation_path=fair-scheduler.xml
> -llama_site_path=llama-site.xml
> We lazily load the configuration for a pool when a query is submitted to that
> pool. This is probably not a good time to fail the server.
> In addition a possibly dangerous configuration for a pool in the XML files
> might never actually be used.
> Add a configuration verification step at server startup. This will log
> warning messages if there is an inconsistency in the configuration.
> There could be multiple checks, for example we could flag if a Root level
> User Quota value is less than a pool level User Quota.
> We could also detect duplicate entries in the configuration.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
