(fix) TWDT / deadlock when sending chunk response where chunk are lon…

[mathieucarbou]

…g to compute (like SD Card reading)

Fix from @Levak
Ref: mathieucarbou/ESPAsyncWebServer#165 (comment)

[Levak]

Well, you were faster than me :)
I was trying to test with the full application with SD, just found few bugs in it (not AsyncTCP, don't worry) :D
For instance, I am getting another Watchdog timeout BUT NO WORRY, this time, I sent 2 parallel SD listing of 700ish files and I see the time of the handler is rising up to 5 seconds before crashing. I would assume it's reading the SD that is that slow, for now.

[mathieucarbou]

Well, you were faster than me :) I was trying to test with the full application with SD, just found few bugs in it (not AsyncTCP, don't worry) :D For instance, I am getting another Watchdog timeout BUT NO WORRY, this time, I sent 2 parallel SD listing of 700ish files and I see the time of the handler is rising up to 5 seconds before crashing. I would assume it's reading the SD that is that slow, for now.

Ah I was going to leave a message in the existing discussion. I will lock it in order to not spread ourselves every where ;-)

I've opened it to run some perf test and even if SSE is 12% faster, HTTP does not support concurrent requests anymore with the fix. I don't know why yet but spawning 10 workers with autocanon for 20 sec, I only reach 1 req / sec.

[mathieucarbou]

Discussion here: mathieucarbou/ESPAsyncWebServer#165