Memory leak in HTTP connection handling about webdriveragent

[qmfrederik]

The Debug Memory Graph raises warnings about 6 leaked types, of which OS_dispatch_queue_serial and OS_dispatch_source seem to be the most interesting. Printing their description yields:

Printing description of $7:
<OS_dispatch_source:
source[0x281400180] = 
{ 
  xref = 1,
  ref = 1,
  target = GCDAsyncSocket[0x28143c580],
  ident = 0xffffffff,
  mask = 0x0,
  pending_data = 0x0,
  registered = 0,
  armed = 0,
  cancelled,
  deleted,
  timer = { target = 0x6963374e1c, deadline = 0x6963374e1c, interval = 0x7fffffffffffffff, flags = 0x0 },
  kevent = 0x281400100,
  filter = timer
}>

[mykola-mokhnach]

This is our fork of RoutingHTTPServer lib: https://github.com/appium/RoutingHTTPServer. Feel free to debug it and propose your fixes.

[mykola-mokhnach]

This also might be related to appium/appium#14416, which was fixed also on the client side in appium@beta. Worth checking anyway

[qmfrederik]

Thanks, I had a look at appium/appium#14416, it seems unrelated at first sight.

I can reproduce the issue by simply initiating a TCP connection (e.g. echo exit | telnet localhost 8100). Doing this in a loop leaks memory in the WDA process. This test never establishes a valid HTTP connection, so much of the middleware of RoutingHttpServer is never invoked.

It makes me think the issue is rather at the CocoaHTTPServer (also forked by appium) or at the GCDAsyncSocket level. For example, the CocoaAsyncSocket source files used by CocaoHTTPServer haven't been updated in 7 years.

I'll have a look and let you know what I find. On the other hand, can you reproduce this problem on your machine? Would be good to know I'm not the only one seeing this :).

[mykola-mokhnach]

Closed as fixed

[nanoscopic]

In what way is this fixed? I continue to see related issues with the HTTP server used.

[qmfrederik]

I could no longer reproduce the issue described in this ticket after #348 and #349 got merged.

Are you seeing you still see this issue (memory leak by just running echo exit | telnet localhost 8100 or curl -X GET http://localhost:8100/health in a loop)?

[nanoscopic]

As shared in other thread, the built in HTTP server is 1.x only, and requires Keep-Alive header as a result to re-use connections. This resolves the issue I was seeing.