pauseTimeout does not work

[joeyh]

"I don't think pauseTimeout works as intended. It sets the pauseTimeoutKey value to Paused, which is promptly lost on resume and tickle, which occur outside the callers control." -- @anacrolix in #351 (comment)

I can confirm, I tried to use pauseTimeout and am still getting timeouts.

warp-3.4.9
time-manager-0.2.4

[kazu-yamamoto]

#351 is too large. Is a remained is only pauseTimeout? If so, would you show us an example code to demonstrate how to reproduce it?

[joeyh]

In my case I simply have a sever that can block for > 30 seconds before responding to a Request. That inactivity triggers slowloris attack protection. I can work up a test case if needed.

[kazu-yamamoto]

Test cases would be very helpful.

[Vlix]

Could you also test if this fails on warp >= 3.4.10 using timeout-manager >= 0.3.1?

[joeyh]

My server uses servant, and is responding with a SourceT IO ByteString consisting of many chunks. After trying for a simpler test case that just delays a response, and finding slowloris protection doesn't happen in that case at all, I think the streaming bytestring response will be needed in the test case unfortunately.

Looking in warp, sendFragment resumes the timeout. Which is outside the caller's control.

[joeyh]

Coming at this from the bottom up, this program shows that time-manager-0.3.1.1, treats a pause followed by a resume as re-enabling the timeout.

#!/usr/bin/env stack
-- stack script --snapshot lts-24.26 --package time-manager --extra-dep time-manager-0.3.1.1

import System.TimeManager
import Control.Concurrent

main = withManager 1000000 $ \mgr ->
	withHandle mgr (print "timeout reached") $ \h -> do
		pause h
		print "pausing for 1 second"
		threadDelay 1000000
		print "resuming"
		resume h
		threadDelay 3000000
		print "end"

==>

joey@darkstar:~>./timemanager.hs
"pausing for 1 second"
"resuming"
"timeout reached"
"end"

This is the same behavior as time-manager-0.2.4 (except that one seems to take a little bit longer to timeout after resume).

[kazu-yamamoto]

I think this is expected behavior.
@joeyh I don't understand what you expect.
Am I missing something?

[Vlix]

In my case I simply have a sever that can block for > 30 seconds before responding to a Request. That inactivity triggers slowloris attack protection. I can work up a test case if needed.

@joeyh Could you please show with a code snippet how you're using pauseTimeout and what you are expecting it to do.

[Vlix]

After trying for a simpler test case that just delays a response, and finding slowloris protection doesn't happen in that case at all.

Slow loris attacks are a client to server attack, not the other way around. This means that the slow loris protection is only active when the server is receiving from or sending to the client. When the request is fully received, the server isn't waiting on the client anymore, so there's no "attack" possibility.
This is why the timeout isn't triggered if you just wait to send a response.