fix(cachingFilter: Allow disabling the content caching filter

[jasonmcintosh]

We see this caching filter break things BADLY. Ability to disable it, defaults to leaving it on.

[jvz]

Works for me, though I'd also be in favor of removing it entirely.

[jasonmcintosh]

:P this way I can backport it ;) since feature flagged

[jasonmcintosh]

@Mergifyio backport release-1.30.x release-1.31.x release-1.32.x

[mergify]

backport release-1.30.x release-1.31.x release-1.32.x

✅ Backports have been created

• #1700 fix(cachingFilter: Allow disabling the content caching filter (backport #1699) has been created for branch release-1.30.x
• #1701 fix(cachingFilter: Allow disabling the content caching filter (backport #1699) has been created for branch release-1.31.x
• #1702 fix(cachingFilter: Allow disabling the content caching filter (backport #1699) has been created for branch release-1.32.x

[jasonmcintosh]

https://github.com/Mergifyio backport release-1.28.x

[mergify]

backport release-1.28.x

✅ Backports have been created

• #1703 fix(cachingFilter: Allow disabling the content caching filter (backport #1699) has been created for branch release-1.28.x

[mattgogerly]

@jasonmcintosh do you have any examples of what this breaks? Curious if any issues we see tally.

[jasonmcintosh]

The biggest problem I've PERSONALLY seen right is situations where gate just fails BADLY on these responses on LARGE pipelines. We've got traces showing orca respond in 2 seconds, then gate takes 80 seconds to do this. It's ENTIRELY in this filter eating up response times. SO there's some of this that's... potentially "scaling" issues or similar on gate to match orca capacities, but it's definitely been painful memory usage as well because of how this operates.

There are other areas this breaks when using different UI libs as well. @jvz can probably talk more to some of those impacts.

[dbyron-sf]

Plus also, why not backport to 1.29/1.30/1.31/1.32?

[jasonmcintosh]

Cause I missed that in my copy/paste ;)
/**

• This filter simply buffers/caches the response so that the Content-Length header can be set.
• Setting the Content-Length header prevents a response from being transferred with chunked
• encoding which may be problematic for some http clients.
  */
  See the above note in the content caching filter itself btw.

[jasonmcintosh]

https://github.com/Mergifyio backport release-1.29.x

[mergify]

backport release-1.29.x

✅ Backports have been created

• #1704 fix(cachingFilter: Allow disabling the content caching filter (backport #1699) has been created for branch release-1.29.x

[jasonmcintosh]

You can see perf data here for example...

[jasonmcintosh]

The original theory on this I think was eTag support. BUT there's some major gotchas around that.
spring-projects/spring-framework#18923 (comment) also an interesting comment :)

[jvz]

The caching filter breaks async controllers that use reactive streams or coroutines for one.