Respect scaling_reserved_ram feature flag - #1760
nickrolfe merged 3 commits into
Conversation
Sorry, something went wrong.
henrymercer
left a comment
There was a problem hiding this comment.
Looks good. It's perhaps a little cautious, but we might want to add an integration test that runs the CodeQL Action with this feature enabled. To do so, we could make a copy of pr-checks/checks/multi-language-autodetect.yml, enable the feature flag via the environment variable, and remove the language autodetection checks.
Sorry, something went wrong.
|
Oh, do add a changelog note too! The one from 24th May is a good example. |
Sorry, something went wrong.
The amount of RAM given to the CodeQL evaluator is the machine's total memory size, minus a reserved amount. Currently, the reserved amount is fixed at 1 GB (or 1.5 GB on Windows). When the scaling_reserved_ram feature flag is enabled, we also add 2% of the total memory size to the reserved amount. This allows for the fact that the kernel will consume more RAM (e.g. for page tables) on machines with more physical RAM.
|
I see the new integration test is running with --ram=5769, while the multi-language autodetection test runs with --ram=5769, so I think it's working. |
Sorry, something went wrong.
There was a problem hiding this comment.
Excellent! (I think you meant the multi-language autodetection test runs with --ram=5907.)
Sorry, something went wrong.
|
Whoops, yes, that's what I meant. |
Sorry, something went wrong.
The amount of RAM given to the CodeQL evaluator is the machine's total memory size, minus a reserved amount.
Currently, the reserved amount is fixed at 1 GB (or 1.5 GB on Windows). When the scaling_reserved_ram feature flag is enabled, we also add 2% of the total memory size to the reserved amount. This allows for the fact that the kernel will consume more RAM (e.g. for page tables) on machines with more physical RAM, so we should see fewer analyses getting killed by the OOM killer.
N.B. This is my first time working on the Action, or even writing TypeScript. I hope my uses of async/await make sense.
Merge / deployment checklist