9.0.x Crash in BaseLogFile

[bryancall]

We are seeing a coredump in production on our 9.0.x branch. We are running an older branch of 9.0.x that was merged on 4/17 (6f3fe28):

[ 0 ] BaseLogFile::roll
  this = 0
[ 1 ] Diags::should_roll_diagslog
[ 2 ] DiagsLogContinuation::periodic
[ 3 ] EThread::process_event
[ 4 ] EThread::execute_regular
[ 5 ] execute
[ 6 ] EThread::execute
[ 7 ] spawn_thread_internal

[bryancall]

[?] Thread 31231 (faulted)
*  [  0] libtscore.so.9.0.0        BaseLogFile::roll() (../../../../../../_vcs/trafficserver9/src/tscore/BaseLogFile.cc:225)
        => add %al, (%rax)
	-----------------------------------------------------------------------
	Signal: [11] SIGSEGV (segmentation violation)
	Reason: SI_TKILL (tkill or tgkill)
	Signal address: 0x63000079f0
	-----------------------------------------------------------------------
    (this) = 0
    start = -- [no location entry found: 0]
    now = 1595107111
   [  1] libtscore.so.9.0.0        Diags::should_roll_diagslog() (../../../../../../_vcs/trafficserver9/src/tscore/Diags.cc:582)
   [  2] traffic_server            DiagsLogContinuation::periodic(int, Event*) (../../../../../_vcs/trafficserver9/src/traffic_server/traffic_server.cc:392)
    (this) = -- [failed to resolve caller frame: 0]
    ((null)) = 597850464
    ((null)) = 0
    diags_log_roll_int = 86400
    diags_log_roll_size = 6000
    diags_log_roll_enable = -- [no location entry found: 0]
    __FUNCTION__ = string(0x77e328, [periodic])
   [  3] traffic_server            EThread::process_event(Event*, int) (../../../../../../_vcs/trafficserver9/iocore/eventsystem/I_Continuation.h:167)
   [  4] traffic_server            EThread::execute_regular() (../../../../../../_vcs/trafficserver9/iocore/eventsystem/UnixEThread.cc:241)
    (this) = 0x2aeac359a240

[bneradt]

I'm 99% certain this is an issue specific to our production boxes related to my SIGUSR2 changes to support third party logrotate tools. That patch has not gone into open source yet because I wanted that to run on our boxes first to make sure it doesn't have issues before publishing it more widely. I've created a local patch to address the crash reported in this issue and I'll report back after that runs a week on our production systems. If there are no longer problems during that time, then we should close this issue because we'll know it's only an issue with our local patch and not a problem in open source.