Watchdog not reset in time about lws-esp32-factory

[lws-team]

Please confirm this still exists on the current -test-server-demos

[FredrikFornstad]

I have not been able to trigger the watchdog with two broswers connected.

However, during startup of esp32 and only one browser connected this procedure seem to be possible to reproduce every time::

1. One Edge-browser connected
2. Restarted esp32 by issuing a "make monitor" command (that boot into lws-test-server)
3. About 10 seconds after esp32 got an IP and the setup of the connection to the browser was ongoing I got a watchdog trigger:

?[35;1m[2018/02/19 17:50:46:6397] NOTICE: ssl_pm_handshakke
?[35;1m[2018/02/19 17:50:46:6422] NOTICE: mbedtls_handshake: ssl ret -0 state 1
?[35;1m[2018/02/19 17:50:46:6532] NOTICE: mbedtls_handshake: ssl ret -0 state 2
?[35;1m[2018/02/19 17:50:46:6620] NOTICE: mbedtls_handshake: ssl ret -0 state 3
?[35;1m[2018/02/19 17:50:46:6703] NOTICE: mbedtls_handshake: ssl ret -0 state 4
Task watchdog got triggered. The followingg tasks did not reset the watchdog in time:

• IDLE (CPU 0)
  Tasks currently running:
  CPU 0: wifi

CPU 1: IDLE
?[35;1m[2018/02/19 17:50:47:7509] NOTICE: mbedtls_handshake: ssl ret -0 state 5
?[35;1m[2018/02/19 17:50:47:7569] NOTICE: mbedtls_handshake: ssl ret -0 state 6

I guess the startup is is an extra stressful moment for the esp32, so maybe not so strange it is happening here. Not sure if it is possible to add an extra yield-statement or two during the setup-phase. If not I guess this can be left without action as it seems Everything continues to startup the Connection anyway.

[lws-team]

Yes... this is something more reasonable than whatever the original thing was. It is calling into mbedtls and sitting there doing the crypto for the best part of a second. The watchdog is right we blocked everything, but it's not because something failed, it's the normal operation and just something cosmetic.

[FredrikFornstad]

Before I close this issue: I connected two browsers (one Edge and the other Firefox on Windows10) to esp32 running lws-test-demo (yes I know, that issue should not be listed here....) and I noticed this:

With both browsers on the "dumb increment counter" page, when I press "Ctrl-F5" in the Edge browser I can see one watchdog warning (the counting goes quite slow on FireFox as expected).
However, if I reload the page in the Firefox browser, there will NEVER be any watchdog warning. The counting slowdown in the Edge browser is similar as I could see in the FireFox browser in the first case.

So the mbedtls is executing differently depending on type of browser.

[lws-team]

Dunno... this watchdog is just there to tell us the core didn't return to the RTOS event loop for some time. For some things we do like mbedtls handshake computation or flash erase, it's true it's possible we just sit there for over a second. It's not an error and it doesn't create problems.

Tls negotiates some parameters on each side, it's not an error if different browsers negotiated different crypto either. So I don't see what to do about that except maybe increase the watchdog timeout.