varnish/libvmod-cookie

[4.0] Assert error in vmod_clean()

Closed this issue · 5 comments

First, thanks for this amazing plugin!

Some times I have this Panic message:

Assert error in vmod_clean(), vmod_cookie.c line 233:
  Condition((cookie)->magic == 0x3BB41543) not true.
thread = (cache-worker)
version = varnish-4.0.3 revision b8c4a34
ident = Linux,3.18.5-tinycore64,x86_64,-smalloc,-smalloc,-hcritbit,epoll
Backtrace:
  0x434eb4: varnishd() [0x434eb4]
  0x7f52e91fcb6b: /usr/local/lib/varnish/vmods/libvmod_cookie.so(vmod_clean+0xab) [0x7f52e91fcb6b]
  0x7f52e91fcc1b: /usr/local/lib/varnish/vmods/libvmod_cookie.so(vmod_parse+0x3b) [0x7f52e91fcc1b]
  0x7f52ea6d3be4: ./vcl.gY_4FnB8.so(VGC_function_vcl_recv+0x274) [0x7f52ea6d3be4]
  0x44178c: varnishd() [0x44178c]
  0x441c65: varnishd(VCL_recv_method+0x45) [0x441c65]
  0x439c18: varnishd(CNT_Request+0x918) [0x439c18]
  0x42f1db: varnishd(HTTP1_Session+0x78b) [0x42f1db]
  0x43d308: varnishd() [0x43d308]
  0x437d61: varnishd(Pool_Work_Thread+0x381) [0x437d61]
req = 0x7f52e342b020 {
  sp = 0x7f52e340e3a0, vxid = 1073807417,  step = R_STP_RECV,
  req_body = R_BODY_NONE,
  restarts = 0, esi_level = 0,
  sp = 0x7f52e340e3a0 {
    fd = 17, vxid = 65539,
    client = 192.168.56.3 60774,
    step = S_STP_WORKING,
  },
  worker = 0x7f52e6163c40 {
    ws = 0x7f52e6163e60 {
      id = "wrk",
      {s,f,r,e} = {0x7f52e6163440,0x7f52e6163440,(nil),+2048},
    },
  VCL::method = RECV,
  VCL::return = abandon,
  },
  ws = 0x7f52e342b1b8 {
    id = "req",
    {s,f,r,e} = {0x7f52e342d010,+288,(nil),+57360},
  }

Maybe I can give some more details if you need.
Thanks.

Are you using the latest version from git master?

Are you seeing multiple ones like this, or different panics?

I compiled it at commit b1fc260.

I have multiple ones like this one, always the same one.

Hi. Please try the 4.0 branch now. (3a2a9ae).

Hi. Did this fix your issue?
(timeout after july 3rd)

Sorry, I will not have enough time to make any test (and I don't work on the related project anymore).
As other tickets on the same subject have been closed, I considered than the bug had been fixed.
Thanks.