mirror of
https://github.com/elastic/kibana.git
synced 2025-06-28 11:05:39 -04:00
1f3426942c
Adds routes to run v8 profiling tools, when running the examples plugins via `--run-examples` See the included README.md for more info
112 lines
4.4 KiB
Markdown
112 lines
4.4 KiB
Markdown
# V8 profiler examples
|
|
|
|
Provides access to the V8 CPU Profiler and Heap Profiler, to inspect
|
|
the Kibana running this plugin.
|
|
|
|
The endpoints are:
|
|
|
|
/_dev/cpu_profile?duration=(seconds)&interval=(microseconds)
|
|
/_dev/heap_profile?duration=(seconds)&interval=(bytes)
|
|
|
|
The default duration is 5 seconds.
|
|
|
|
For more information on these V8 APIs, see:
|
|
|
|
- https://chromedevtools.github.io/devtools-protocol/v8/Profiler/
|
|
- https://chromedevtools.github.io/devtools-protocol/v8/HeapProfiler/
|
|
|
|
Note that due to bugs or limitations,
|
|
it's not possible to generate heap snapshots using the techniques used
|
|
to generate the cpu and heap profiles.
|
|
|
|
Try them right now, assuming you started Kibana with `--run-examples`!
|
|
|
|
- [`http://localhost:5601/_dev/cpu_profile`](http://localhost:5601/_dev/cpu_profile)
|
|
- [`http://localhost:5601/_dev/heap_profile`](http://localhost:5601/_dev/heap_profile)
|
|
|
|
|
|
When using curl, you can use the `-kOJ` options, which:
|
|
|
|
- `-k --insecure`: allow HTTPS usage with self-signed certs
|
|
- `-O --remote-name`: use the server-specified name for this download
|
|
- `-J --remote-header-name`: use the `Content-Disposition` as the name of
|
|
the download
|
|
|
|
So one of the following should work for you, to run a 10 second CPU profile
|
|
using an interval of 100μs (default: 5s / 1000μs):
|
|
|
|
```
|
|
curl -OJ "http://elastic:changeme@localhost:5601/_dev/cpu_profile?duration=10&interval=100"
|
|
curl -kOJ "https://elastic:changeme@localhost:5601/_dev/cpu_profile?duration=10&interval=100"
|
|
```
|
|
|
|
The files generated will be:
|
|
|
|
MM-DD_hh-mm-ss.cpuprofile
|
|
MM-DD_hh-mm-ss.heapprofile
|
|
|
|
These filetypes are the ones expected by various V8 tools that can read these.
|
|
|
|
You can use these URLs in your browser, and the files will be saved with the
|
|
generated names.
|
|
|
|
## profile / heap profile readers
|
|
|
|
The traditional tools used to view these are part of Chrome Dev Tools (CDT)
|
|
and now VS Code also supports viewing these files. They provide
|
|
similar capabilities.
|
|
|
|
There doesn't seem to be a much doc available on how to use the viewers
|
|
for these files. The Chrome Dev Tools docs are extremely old and appear
|
|
to be out-of-date with the current user interface. The VS Code
|
|
documentation [Analyzing a profile][] is more recent, but there's not
|
|
much there.
|
|
|
|
[Analyzing a profile]: https://code.visualstudio.com/docs/nodejs/profiling#_analyzing-a-profile
|
|
|
|
For CPU profiles, open CDT and then click on the "Performance" tab. You
|
|
should be able to drop a file right from Finder / Explorer onto the CDT
|
|
window, and then get the visualization of the profile. If you
|
|
downloaded the profile right from the browser, using the URL in the URL
|
|
bar, you can drop the download file from the download status button
|
|
right into CDT.
|
|
|
|
For heap profiles, open CDT and then click on the "Memory" tab. Drag and
|
|
drop doesn't seem to work here, but you can load a file via a file
|
|
prompter by clicking the "Load" button at the bottom of the "Memory"
|
|
pane. You will probably need to scroll to see the button.
|
|
|
|
VSCode now supports `.cpuprofile` files and `.heapprofile` files
|
|
directly, displaying them as a table of function timings. There is also
|
|
[an extension available to display flame charts][] installed by clicking
|
|
on the grey-ed out "flame" button on the top-right of the cpu profile
|
|
view.
|
|
|
|
[an extension available to display flame charts]: https://marketplace.visualstudio.com/items?itemName=ms-vscode.vscode-js-profile-flame
|
|
|
|
There seem to be problems with both CDT and the VS Code tools, at times.
|
|
The flame charts in VS Code seem to go haywire sometimes. The heap
|
|
profile tables in CDT don't expand. Etc. So, beware, and be prepared to
|
|
have to use multiple tools to analyze these files.
|
|
|
|
An alternate view of CPU profiles, which organizes files based on
|
|
"packages", is available with the **NO**de **PRO**filer (`no-pro`) thing
|
|
available at https://pmuellr.github.io/no-pro/ . It also supports
|
|
drag-n-drop of CPU profile files. Note that you can get more
|
|
directories to show up as "packages", by bringing up CDT and running the
|
|
following code:
|
|
|
|
localStorage['fake-packages-dirs'] = "x-pack/plugins,packages,src/core"
|
|
|
|
|
|
## tips / tricks
|
|
|
|
If you're handy with Mac Finder, or other ways of auto-launching apps
|
|
based on file extensions, it's easy to associate `.cpuprofile` files
|
|
with vscode.
|
|
|
|
Since the http endpoints are GET requests, they are easy to bookmark.
|
|
Start a profile by clicking on a bookmark.
|
|
|
|
When these files are downloaded via Chrome, you can typically launch
|
|
them directly from the download bar, or drag the file to a viewer.
|