Scroll down to the script below, click on any sentence (including terminal blocks!) to jump to that spot in the video!
We know that the scenario will be executed against our production server only. If we profiled a local page, this stuff has no effect. That means that the results of these profiles should have less variability. Not no variability: if your production server is under heavy traffic, the profiles might be slower than normal. But, it will have less variability than trying to compare a profile that you created on your local machine with a profile created on production: those are totally different machines and setups.
I also recommend adding
samples 10 to each scenario. This will then use
10 samples (like normal Blackfire profiles) and further reduce variability:
visit url("/") name "Homepage" samples 10 ...
This means that you can... maybe add some time-based assertions... as long as you're conservative. For example, on the homepage, let's
main.wall_time < 100ms.
By the way, most metrics start with
metrics. and you can look on the timeline to see what's available. A few metrics - like wall time and peak memory - start with
Anyways, as you can see inside Blackfire, our homepage on production normally has a wall time of about 50ms... so 100ms is fairly conservative. But time-based metrics are still fragile. Doing this will likely result in some random failures from time-to-time.
Let's commit this:
git add .git commit -m "adding homepage time assertions"
symfony deploy --bypass-checks
While that's deploying, I want to show you a super powerful feature that we won't have time to experiment with: custom metrics. Google for "Blackfire metrics". In addition to the timeline, this page also lists all of the metrics that are available.
But you can also create your own metrics inside
.blackfire.yaml. In addition to
scenarios, we can have a
metrics key. For example, this creates a custom metric called "Markdown to HTML". The real magic is the
matching_calls config: any time the
toHtml method of this made-up
Markdown class is called, its data will be grouped into the
That's powerful because you can immediately use that metric in your tests. For example, you could assert that this metric is called exactly zero times - as a way to make sure that some caching system is avoiding the need for this to ever happen on production. Or, you could check the memory usage... or other dimension.
You can use some pretty serious logic to create these metrics: making it match only a specific caller for a function, OR logic, regex matching and ways to match methods, calls from classes that implement an interface and many other things. You can even create separate metrics for the same method based on which arguments are passed to them. They went a little nuts.
Anyways, let's check on the deploy. Done! Go back - I'll close this tab - and let's create a new build. Call it "With homepage wall time assert". Start build!
And... it passes! This time we can see an extra constraint on the homepage: wall time needs to be less than 100ms. If it's greater than 100ms and you have notifications configured, you'll know immediately.
Next: now that we have this idea of builds being created every 6 hours, we can do some cool stuff, like comparing a build to the build that happened before it. Heck we can even write assertions about this! Want a build to fail if a page is 30% slower than the build before it? We can do that.