px/perf_flamegraph distortions

[sync-by-unito[bot]]

If running the px/perf_flamegraph script, it's possible more than 10,000 stack traces are queried. Our scripts, however, silently cap this to 10,000 rows, which results in distorted flamegraphs. The distorted flamegraphs are then very misleading.

Note that this happens when the (time window * number of pods) is large.

We need a way to either avoid these distortions or detect them and show an error message with actions the user can take to rectify the situtation.

┆Issue is synchronized with this Jira Bug by Unito

[sync-by-unito[bot]]

➤ Vihang Mehta commented:

I started investigating this and it’s actually quite difficult to solve with our current limitations. Here are my discoveries, https://www.notion.so/pixielabs/Accurate-Flame-Graphs-in-Pixie-1014d9e36ec749e98563c3ad652cc721 ( https://www.notion.so/pixielabs/Accurate-Flame-Graphs-in-Pixie-1014d9e36ec749e98563c3ad652cc721 ) going to get some input from Phillip Kuznetsov Natalie Serrino Zain Asgar to see if there are any other possible solutions I missed.