-
Notifications
You must be signed in to change notification settings - Fork 505
Closed
Description
Issue Type:
- Bug report
- Feature request
Do you think it makes sense to revise the necessity or behavior of maxRequestKB
and maxResponseKB
after #2189 ?
I don't know the initial motivation behind those, but I guess that it may be to protect linkerd from OOMs.
To me it would make sense to have following set of properties:
streamAfterContentLengthKB
- current behavior. Effectively defines how much space in heap large unchunked request can take at mostmaxRequestKB
,maxResponseKB
- max request/response size, both chunked and unchunked. Having it only for unchunked requests as it is now does not have much sense, since we already havestreamAfterContentLengthKB
(unless of course the motivation behind those properties is something else than preventing OOM). Having these limits for all requests/responses on the other hand, makes total sense for me: we may want to protect services in our service mesh from unreasonably large requests without polluting services logic.
However in this case, I believe, we should make it unlimited by default (right now finagle allows 2 GiB max, but I think we can try discussing this with finagle team).
Metadata
Metadata
Assignees
Labels
No labels