-
Notifications
You must be signed in to change notification settings - Fork 62
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
Performance degradation when updating from 10.17.0.7 to 10.19.0.9 #859
Comments
Hi @adri! Thank you for bringing this to the team’s attention. Could you contact New Relic support by following these steps. This way we will be able to gather additional details about the issue at hand. If your subscription level does not include technical support support plan, the other option available to customers with free plans is to use New Relic’s community support channel: Explorers Hub. For all available options on how to get support, please see these resources. |
@newrelic-php-agent-bot What does that performance degradation has to do with our support plan. We are paying customers. We report that your latest SDK degrades performance. And the previous did not. Clearly something is off. |
@ruudk thanks for the question, and we'll try to clear up the intent. The performance degradation has nothing to do with the support plan and is an issue we are investigating. Sometimes (as in this case), in order to investigate further, we need certain details from customers that cannot be shared in a public repo. Our method for collecting this data in a manner to provide privacy and security to the customer is via our support channel which can be reached in various ways (as detailed in the bot message). |
We noticed this when our systems were upgraded from v10.17.0.7 to v10.18.0.8, our endpoints were performing dramatically worse, even basic nothing endpoints like health checks. We build our base docker containers with the latest versions nightly so within a few days of the v10.18.0.8 release it was very clear something had gone wrong. This is a health check endpoint that does nothing but return a string This is a more involved endpoint which is usually consistently <60ms Given the change log, we thought perhaps it was due to JIT finally being available.
We have since tested with many different configurations of JIT, including JIT off and it seems to us at least this performance regression is due v10.18.0.8 itself and not JIT. It seems the same issue is also present in v10.19.0.9 We've rolled back and pinned to v10.17.0.7 and are watching this issue & release announcements hoping for a fix, just commenting here to give it weight. If it helps, our newrelic.ini has long had the transaction_tracer disabled to minimise performance impacts
|
We noticed the same as soon as we merged the new version, we had to roll it back to |
@slt @ebeltramo96 If you follow the bot instructions, our support team could work with you to collect additional information such as debug logs and other sensitive system info in a secure manner. |
10.20.0.10 has been released: Anyone knows if this bug still occurs with that version? |
OS: Debian 12 bookworm @zsistla I just created a Case #00224017 |
As @slt pointed out, release 10.18.0.8 added support for PHP’s Just-In-Time Compilation. This support required changes to how New Relic PHP Agent hooks into Zend Engine. I.e. for PHPs 8.0+ it no longer uses the method of overriding zend_execute hook but it uses observer API instead. This change came with an increased overhead. New Relic PHP Engineering team is looking into options to reduce the overhead when observer API is used. |
thx for looking into this. This is blocking us from using distributed tracing, so we hope this is fixed soon. |
Description
When updating from 10.17.0.7 to 10.19.0.9 we saw an average response increase of ~10ms, from ~37ms to ~47ms. After reverting the response time decreased.
Steps to Reproduce
I don't know
Expected Behavior
No performance impact
Your Environment
Symfony v7.0
PHP 8.3.2
Debian 12
The text was updated successfully, but these errors were encountered: