utl: Adds live prometheus monitoring to OpenROAD #6741
Conversation
OpenROAD includes a metrics endpoint server that can track internal tool metrics over time. To use this feature you need to do the following start the prometheus and grafana collectors [Detailed instructions](/etc/monitoring/README.md): ```shell $ cd etc/monitoring $ docker compose up -d ``` This will start a grafana endpoint ready to collect from the OpenROAD application you would like to track. By default it's looking for an http server running on port 8080 on your localhost. To start the metrics endpoint in OpenROAD, run: ```tcl utl::startPrometheusEndpoint 8080 ``` This is all configurable in the docker compose file, and you should be able to access grafana by going to http://localhost:3000 username: admin, password: grafana. Go to the dashboard tab and click service, then OpenROAD to see the pre-made dashboard. Signed-off-by: Ethan Mahintorabi <[email protected]>
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
|
This appears to have a copy of https://github.com/jupp0r/prometheus-cpp. It should be in third-party. |
|
This is a proper fork. I think it should live in utl. I renamed all the namespaces and redid the imports so it could be a proper part of utl. |
Signed-off-by: Ethan Mahintorabi <[email protected]>
|
Is there a reason we want a fork? Will we never want upstream updates? |
|
The library hasn't been updated in two years and is arguably complete. At the end of the day it's essentially just a map with some keys and values and a way to serialize it into a text file. I have more ideas on how to drive it forward that are good for OpenROAD, but arguably not in the interest of the original library. Like adding more pluggable inheritance and a way to stub in Google's monitoring backend Monarch. Even in this version I removed its dependency on its original http library. Moved to pull vs push, and implemented the http stuff based on boost |
|
Also this is the original implementation not the one you linked to. https://github.com/biaks/prometheus-cpp-lite |
Signed-off-by: Ethan Mahintorabi <[email protected]>
|
The one I pointed probably started as a copy but has more recent commits and many more stars and forks. It seems like the original died but this one is still alive fwiw. |
|
That's the official otel one. We just don't really need what it provides, namely a bunch of http server stuff, and unnecessary performance cruft. It's better to just have our own fork that we have full control over. Performance stuff matters in webservers where it could conceivably add to response latency, but we're talking 100s of milliseconds at worst, and when you're talking about OpenROAD we're talking hours of runtime against negligible metrics cost. |
| | Switch Name | Description | | ||
| | ----- | ----- | | ||
| | `name` | Name of the command/message to query. | | ||
| | `-manpath` | Include optional path to man pages (e.g. ~/OpenROAD/docs/cat). | | ||
| | Switch Name | Description | | ||
| | ----------- | ------------------------------------------------------------------------------------------------------------------------------------ | | ||
| | `name` | Name of the command/message to query. | | ||
| | `-manpath` | Include optional path to man pages (e.g. ~/OpenROAD/docs/cat). | |
There was a problem hiding this comment.
Why is this reformatted with extraneous whitespace?
| @@ -59,6 +59,9 @@ | |||
|
|
|||
| namespace utl { | |||
|
|
|||
| class PrometheusMetricsServer; | |||
| class Registry; | |||
There was a problem hiding this comment.
Registry is a very generic name. I would prefer to see the code the prometheus subdir in namespace like utl::prometheus
| class PrometheusMetricsServer | ||
| { | ||
| std::thread worker_thread_; |
There was a problem hiding this comment.
The private section goes last https://google.github.io/styleguide/cppguide.html#Declaration_Order
|
|
||
| public: | ||
| ~PrometheusMetricsServer(); | ||
| PrometheusMetricsServer(std::shared_ptr<Registry>& registry_, |
There was a problem hiding this comment.
Don't inline; ctor before dtor https://google.github.io/styleguide/cppguide.html#Declaration_Order
| worker_thread_.join(); | ||
| } | ||
|
|
||
| std::string SnapshotPrometheusMetrics(Registry* registry) |
There was a problem hiding this comment.
local functions should be in an anonymous namespace https://google.github.io/styleguide/cppguide.html#Internal_Linkage
OpenROAD now includes a metrics endpoint server that can track internal tool metrics over time.
To use this feature you need to do the following start the prometheus and grafana collectors
Detailed instructions:
$ cd etc/monitoring $ docker compose up -dThis will start a grafana endpoint ready to collect from the OpenROAD application you would like to track. By default it's looking for an http server running on port 8080 on your localhost.
To start the metrics endpoint in OpenROAD, run:
This is all configurable in the docker compose file, and you should be able to access grafana by going to http://localhost:3000 username: admin, password: grafana. Go to the dashboard tab and click service, then OpenROAD to see the pre-made dashboard.