You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Great work y'all are doing here! We've run into a bit of a snag dealing with errors and stack traces.
The problem:
As bundle size increases, the duration of calculating a stack trace with source maps of an error thrown during resolver execution takes more time to compute.
We narrowed down the culprit to the following check:
We did a flame graph for an error thrown and this resulted in the following:
It takes this amount of time regardless of which resolver throws an error.
We currently bundle our server (along with tree-shaken dependencies) in a single file so we can upload it to a lambda. We are at around 76MB for our total bundled file which is not great (😟) but still very usable for us in production.
Our workaround right now is to explicitly set captureStackTrace and bypass having to perform this calculation. This decreases the execution time from 1154ms to 20ms. However, we are left with no stack traces.
Any suggestions or help would be greatly appreciated, thanks!
The text was updated successfully, but these errors were encountered:
So doing Error().stack which is the fallback when not calling captureStackTrace doesn't take up this much compute? We could just always leverage Error().stack and filter out the GraphQLError line which is essentially all that captureStackTrace should be doing here. Alternatively, our stack will be very deep, we could set a stackTraceLimit which should reduce the performance impact, all though the lookup in the sourcemaps should always be expensive.
The node team does have this as an issue as well nodejs/node#41541 (comment) - in case you are leveraging ESBuild there are tips in there as well. Also a noteworthy PR nodejs/node#43875 - so this performance hit should only be applicable when the error is unhandled.
Great work y'all are doing here! We've run into a bit of a snag dealing with errors and stack traces.
The problem:
As bundle size increases, the duration of calculating a stack trace with source maps of an error thrown during resolver execution takes more time to compute.
We narrowed down the culprit to the following check:
graphql-js/src/error/GraphQLError.ts
Lines 144 to 158 in 9c90a23
We did a flame graph for an error thrown and this resulted in the following:
It takes this amount of time regardless of which resolver throws an error.
We currently bundle our server (along with tree-shaken dependencies) in a single file so we can upload it to a lambda. We are at around 76MB for our total bundled file which is not great (😟) but still very usable for us in production.
Our workaround right now is to explicitly set
captureStackTrace
and bypass having to perform this calculation. This decreases the execution time from1154ms
to20ms
. However, we are left with no stack traces.Any suggestions or help would be greatly appreciated, thanks!
The text was updated successfully, but these errors were encountered: