[Blackfire] Unable to run Blackfire profiles due to load balancing

[Hugo Alliaume]

Hi!

We are facing a load-balancing issue when running Blackfire Scenarios on our production application, which has 3 dynos:

<failure>  └ <failure>Profiling progress is inconsistent (progress is going backward). 
That happens for instance when the project's infrastructure is behind a load balancer. 
Please read https://blackfire.io/docs/up-and-running/reverse-proxies#configuration-load-balancer

The Blackfire documentation about load balancer recommends to route all requests containing a header X-Blackfire-Query to the same server, which AFAIK is not possible with Heroku.

Another solution is to configure which socket the Blackfire Agent should listen, e.g. tcp://0.0.0.0:8307, which should be doable by configuring the env var BLACKFIRE_AGENT_SOCKET, but it looks like this env var is hard-coded:

heroku-buildpack-php/support/build/blackfire

Lines 34 to 38 in eaecccf

 tee ${OUT_PREFIX}/bin/export.blackfire.sh > ${OUT_PREFIX}/bin/profile.blackfire.sh <<'EOF' 
 # hard-code these two; no need for users to override them 
 export BLACKFIRE_LOG_FILE=stderr 
 export BLACKFIRE_AGENT_SOCKET="unix:///app/.heroku/php/var/blackfire/run/agent.sock" 
 EOF 

What is the best way to fix this, does someone already has this issue?

Thanks!

[David Zülke]

Hi,

The tcp://0.0.0.0:8307 approach works by having a single Agent on one server (listening on all interfaces due to 0.0.0.0), that all Probes then talk to over the network. That doesn't work on "normal" Heroku, since Dynos are isolated; it would be possible in a Private Space.

However, you should be able to enable https://devcenter.heroku.com/articles/session-affinity for your Heroku app to always route traffic from the same user to the same dyno.