Add memory-profiler and add ability to selectively enable rack-mini-profiler #24022
Conversation
…rofiler on production front ends for admins
|
Unlike adhoc instances, I believe the front end instances don't have public IP addresses. If that's the case, you could use Perhaps you could just keep refreshing your browser until the load balancer sends your request to the server with profiling enabled? |
|
Hold off. This isn't working :( On an adhoc, or in development, setting Initial investigation locally shows that the authorization is being reset between when its enabled and when the page is actually rendered. From some initial logging of stack traces, this appears to be coming from (a conflict with?) the newrelic_rpm gem. I need to investigate / debug further... |
As Suresh noted, our production-frontend EC2 instances don't have public ip addresses for security purposes. You can reach the private IP from within our VPC, either on See Debugging Frontend instances on our Wiki for instructions on setting up SSH port-forwarding through |
| @@ -81,7 +83,7 @@ def reset_session_endpoint | |||
|
|
|||
| def render_404 | |||
| respond_to do |format| | |||
| format.html {render file: 'public/404.html', layout: 'layouts/application', status: :not_found} | |||
| format.html {render file: 'pblic/404.html', layout: 'layouts/application', status: :not_found} | |||
There was a problem hiding this comment.
seems like a typo? (also seems to be the cause of ci test failures)
| # setting the environment variable prior to running the server, or by hitting | ||
| # a page with a "pp=enabled" query string parameter and then refreshing the page. | ||
| Rack::MiniProfiler.config.pre_authorize_cb = lambda {|_env| ENV['RACK_MINI_PROFILER'] == 'on'} | ||
| # Only pre-authorize the min profiler when this setting is enabled (in locals.yml). |
|
Thanks @wjordan for all the help figuring our how to make this work on adhoc (and therefore hopefully production 🤞). The key was whitelisting the cookie that rack-mini-profiler uses internally to maintain state so it is not stripped. |
on production front ends for admins.
We have had the
rack-mini-profileravailable in development mode for several years, though I don't think anyone has been using it. We also had a shortcut in development, where adding?pp=enabledto the query would turn it on for a running server. For simplicity I removed that and consolidated in one code path. It's not a big deal to add that key to locals and restart the server, and it simplifies the production code path.In order for the profiler to be enabled, the following must be met:
CDO.rack_mini_profiler_enabled(from locals.yml for instance) must be set to true.?ppmust be in the query.Relevant options for profiling memory (source):
?pp=profile-memoryto generate a memory report?pp=profile-gcto report garbage collection stats?pp=analyze-memoryto report ObjectSpace statsOnce deployed, I intend to ssh into a single front end, add the
rack_mini_profiler_enabledsetting inlocals.yml, restart the dashboard service, then hit that front end directly as outlined in this comment to get live memory reports. Initially, I'll take a few samples, just after DTP and after it's been running for a while, look for patterns, and report to the team.Additional context and reasoning.
MemoryStore, RedisStore, MemcacheStore, and FileStore. I discussed with @wjordan and decided to use the file store so we can share across processes but not across servers, and because we intend this to be used on demand and in low volume on a single front end.pp=query param, whenCDO.rack_mini_profiler_enabledis set.if falsestatements".See https://github.com/MiniProfiler/rack-mini-profiler for more description and options.