[
https://issues.apache.org/jira/browse/HDDS-17588?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
rajesh updated HDDS-17588:
--------------------------
Description:
Parent task: HDDS-17586.
On an OM write trace, compare each parent span with its children. A parent of
about 100ms whose children are about 1ms each is missing the span for the work
that used the time. Add that span. A span that only repeats its parent does not
count.
Check volume, bucket and key create and commit:
* The request span ({{CreateVolume}}, {{CreateBucket}}, {{CreateKey}},
{{CommitKey}}) against {{OM.preExecute}}, {{OM.applyTransaction}} and the
double-buffer {{DB}} spans.
* {{OM.preExecute}} against the ACL check and, for key create, the SCM
{{allocateBlock}} call.
* {{OM.applyTransaction}} against {{OM.validateAndUpdateCache}}.
One local trace already shows the shape: {{CreateVolume}} was about 45ms, while
{{OM.preExecute}}, {{OM.applyTransaction}} and the {{DB}} spans added up to
about 20ms. Find that missing time and name a span for the step that used it.
was:
Wrap OM components with \{{TracingUtil.createProxy(impl, Interface.class,
SpanKind.INTERNAL)}}, so that every call gets a \{{<ImplClass>.<method>}} span,
for example \{{OMKeyCreateRequest.validateAndUpdateCache}} or
\{{KeyManagerImpl.listStatus}}:
* \{{RequestHandler}} (\{{OzoneManagerRequestHandler}}), in the state machine
and in the server-side translator
* OM client requests, through the new \{{IOMClientRequest}} interface returned
by \{{OzoneManagerRatisUtils#createClientRequest}}
* \{{KeyManager}}, \{{BucketManager}}, \{{VolumeManager}}, \{{S3SecretManager}}
and \{{OMMultiTenantManager}}
* \{{IAccessAuthorizer}}
Trivial or very frequently called methods are excluded with \{{@SkipTracing}}.
A JDK dynamic proxy implements only its interface, so the following refactors
come first (no functional change):
* use the \{{KeyManager}} interface instead of \{{KeyManagerImpl}} casts;
* test read-path datanode sorting through \{{KeyManager#listStatus}};
* move request handler fault injection into a test decorator;
* introduce \{{IOMClientRequest}}.
Summary: Find OM write spans whose children do not account for the
parent time (was: Create tracing proxies for OM components)
> Find OM write spans whose children do not account for the parent time
> ---------------------------------------------------------------------
>
> Key: HDDS-17588
> URL: https://issues.apache.org/jira/browse/HDDS-17588
> Project: Apache Ozone
> Issue Type: Sub-task
> Reporter: rajesh
> Assignee: rajesh
> Priority: Major
>
> Parent task: HDDS-17586.
> On an OM write trace, compare each parent span with its children. A parent of
> about 100ms whose children are about 1ms each is missing the span for the
> work that used the time. Add that span. A span that only repeats its parent
> does not count.
> Check volume, bucket and key create and commit:
> * The request span ({{CreateVolume}}, {{CreateBucket}}, {{CreateKey}},
> {{CommitKey}}) against {{OM.preExecute}}, {{OM.applyTransaction}} and the
> double-buffer {{DB}} spans.
> * {{OM.preExecute}} against the ACL check and, for key create, the SCM
> {{allocateBlock}} call.
> * {{OM.applyTransaction}} against {{OM.validateAndUpdateCache}}.
> One local trace already shows the shape: {{CreateVolume}} was about 45ms,
> while {{OM.preExecute}}, {{OM.applyTransaction}} and the {{DB}} spans added
> up to about 20ms. Find that missing time and name a span for the step that
> used it.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
