-
Notifications
You must be signed in to change notification settings - Fork 834
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Metric exporter REUSABLE_DATA memory mode configuration options #6304
Metric exporter REUSABLE_DATA memory mode configuration options #6304
Conversation
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #6304 +/- ##
============================================
- Coverage 91.11% 91.09% -0.02%
- Complexity 5760 5772 +12
============================================
Files 627 627
Lines 16799 16852 +53
Branches 1718 1720 +2
============================================
+ Hits 15306 15351 +45
- Misses 998 1006 +8
Partials 495 495 ☔ View full report in Codecov by Sentry. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks a lot for taking this. It's on my list, but I haven't gotten around to it yet.
I've added this in the feature implementation plan: #5105 (comment) |
@jack-berg Do we want to add support also |
Yes I think its good to add support to prometheus. Wondering whether we'd want just a single property instead of multiple:
|
I favor your single property suggestion |
…y-java into reuseable-memory-mode-env-var
@asafm I've pushed commits to:
|
exporters/otlp/all/src/main/java/io/opentelemetry/exporter/otlp/internal/OtlpConfigUtil.java
Show resolved
Hide resolved
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the explanations @jack-berg. Looks like a great start to configure a potentially amazing perf improvement.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sorry for the late reply, last week I was busy for a lecture I gave in a conference.
this.prometheusRegistry = builder.prometheusRegistry; | ||
this.otelScopeEnabled = builder.otelScopeEnabled; | ||
this.allowedResourceAttributesFilter = builder.allowedResourceAttributesFilter; | ||
this.executor = builder.executor; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Don't you want to copy the memory mode as well?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Good point
@@ -85,6 +89,11 @@ public AggregationTemporality getAggregationTemporality(InstrumentType instrumen | |||
return prometheusMetricReader.getAggregationTemporality(instrumentType); | |||
} | |||
|
|||
@Override |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The only difference I see in PrometheusHttpServer compared with PeriodMetricReader is that in Prometheus, multiple threads can call the metric producer collect(). I think if the mode is turned at REUSABLE_DATA, we should protect and make it only one request at a time.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There's a lock preventing concurrent reads from a particular meter here, but its not enough: A meter could be read from sequentially, but PrometheusHttpServer might still be serializing the MetricData from the first read when the second read begins and overwrites the data. So yes, a lock is needed when PrometheusHttpServer has reusable data enabled. This is interesting because it makes the tradeoff really clear: Do you want reduced memory allocation but no concurrent reads? Or higher memory allocation w/ concurrent reads?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yes. I think we can place that in the javaDoc of the memory mode setter of the builder, once we fix the bug. Do we have place we write documentation of exporters in general ?
| otel.logs.exporter | OTEL_LOGS_EXPORTER | List of exporters to be used for logging, separated by commas. Default is `otlp`. `none` means no autoconfigured exporter. | | ||
| otel.java.experimental.exporter.memory_mode | OTEL_JAVA_EXPERIMENTAL_EXPORTER_MEMORY_MODE | If `reusable_data`, enable reusable memory mode (on exporters which support it) to reduce allocations. Default is `immutable_data`. This option is experimental and subject to change or removal.**[1]** | | ||
|
||
**[1]**: NOTE: The exporters which adhere |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Would you like me to add documentation to the website about memory mode?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think there's an issue open somewhere to move all this autoconfigure docs to the website. For the moment, opentelemetry.io just links to this page for anything related to SDK environment variables. I think that's fine for now.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I guess I'm asking if we should elaborate more on the memory mode, or the configuration description suffices?
The work that has been done to introduce the
REUSABLE_DATA
memory mode dramatically reduces memory allocations for the metrics SDK. And because the SDK doesn't allow concurrent collections, it should be safe to use in built in exporters like OTLP and prometheus. (The only time it wouldn't be safe to use would be if an exporter held on to MetricData refs after completing the CompletableResultCode returned by export.)This PR adds experimental API to programmatically set reusable data memory mode:
And an experimental system property / env var for setting reusable memory mode:
This will set memory mode for any exporter that supports setting reusable data memory mode. Currently, that include
OtlpHttpMetricExporter
,OtlpGrpcMetricExporter
, andPrometheusHttpServer
.After we incubate this for a bit, we should make these configuration options part of the permanent public surface area, and probably enable REUSABLE_DATA mode by default for OTLP and prometheusmetrics, since its almost always the right thing to do.
cc @asafm