Random server crashed on missing sessions (?)

[mrtopf]

Hi!

I am experiencing random server crashes with the following tracebacks:

^[[31m[2012-09-24 12:37:25.212] [ERROR] console - ^[[39mTypeError: Cannot set property 'padId' of undefined
    at /home/eplite/etherpad-lite/src/node/handler/PadMessageHandler.js:838:37
    at /home/eplite/etherpad-lite/src/node_modules/async/lib/async.js:486:21
    at /home/eplite/etherpad-lite/src/node_modules/async/lib/async.js:185:13
    at /home/eplite/etherpad-lite/src/node_modules/async/lib/async.js:108:13
    at /home/eplite/etherpad-lite/src/node_modules/async/lib/async.js:119:25
    at /home/eplite/etherpad-lite/src/node_modules/async/lib/async.js:187:17
    at /home/eplite/etherpad-lite/src/node_modules/async/lib/async.js:491:34
    at /home/eplite/etherpad-lite/src/node_modules/async/lib/async.js:190:13
    at /home/eplite/etherpad-lite/src/node_modules/async/lib/async.js:94:25
    at /home/eplite/etherpad-lite/src/node_modules/async/lib/async.js:187:17
^[[32m[2012-09-24 12:37:25.212] [INFO] console - ^[[39mgraceful shutdown...
^[[32m[2012-09-24 12:37:25.307] [INFO] socket.io - ^[[39mtransport end (close timeout)

As the code looks like this:

  //Save in sessioninfos that this session belonges to this pad
  var sessionId=String(client.id);
  sessioninfos[sessionId].padId = padIds.padId; // this is 838

it seems that the session is not defined. I am not sure why and wonder how I can find that out. Moreover the question is if this should be more robust then. Unfortunately I understand the code not enough to understand if client.id must actually be valid at this point or not. Could restarts lead to undefined sessions?

Unfortunately I don't see a way right now to reproduce it.

[JohnMcLear]

Which branch are you running?

[marcelklehr]

I ran across that one, too. But, since I was debugging a new feature, I thought that was due to a bug I had introduced.

[mrtopf]

I forked master (and did some changes but in the static/CSS section so not related) and updated on friday.

[marcelklehr]

It's got to be still present in develop. The relevant line in PadMessageHandler.js is 878.

[marcelklehr]

This probably occurs, when a user connects and disconnects right after CLIENT_READY is sent.

While, one server "thread" processes the disconnect and removes the session from sessioninfos, meanwhile another "thread" is processing the CLIENT_READY (which takes much longer) and assumes, that the session is still valid.

Turns out, you do have to make your node programs "thread-safe", after all... (at least, if you use async)

[marcelklehr]

Fixed in #1028. closing...