Re: slow startup after upgrade

[q2t2hzmwuk at snkmail dot com]

Corinna Vinschen corinna-cygwin-at-cygwin.com |cygwin_ml_nodigest| wrote at 22:13 +0100 on Feb 24, 2015:
 > Hi Roger,
 >
 > On Feb 24 19:55, Roger Orr wrote:
 > > Hello Corinna,
 > > It seems slightly faster than the previous patch and I've not noticed a
 > > downside yet.
 > >
 > >
 > > CYGWIN_NT-6.1-WOW LCLDN-DEV24 1.7.35s(0.286/5/3) 20150220 15:47:55 i686
 > > Cygwin:
 > >
 > > ~37ms to run echo.exe from Windows command prompt
 > > ~60ms to run .\id.exe -a from Windows command prompt
 > >
 > > CYGWIN_NT-6.1-WOW LCLDN-DEV24 1.7.35s(0.286/5/3) 20150223 21:02:38 i686
 > > Cygwin:
 > >
 > > ~35ms to run echo.exe from Windows command prompt
 > > ~53ms to run .\id.exe -a from Windows command prompt
 >
 > That's a nice result.
 >
 > However, I don't quite understand this result for the older DLL.
 > Weren't you reporting >4 secs as startup time from 1.7.35?!?
 >
 > On another note:
 >
 > I just uploaded a new developer snapshot (2015-02-24).  This snapshot
 > should improve mkpasswd/mkgroup or, generally speaking, enumerating AD
 > accounts, a lot.  Can you give it a try?
 >
 > While you're at it, does the new snapshot still stop after 3.5K accounts
 > even though you think there are 8K accounts?  If so, I'd be interested
 > to investigate this further.  The reason is, while testing my today's
 > performance improvements, I stumbled over a bug in my code which also
 > resulted in enumerating less accounts as desired.  So I'm not entirely
 > sure your problem isn't related to a bug either.

That mkpasswd symptom was reported by me (well, maybe others, too).
I'm running it again, but it's still very slow, AFAICS.  I'll report
how long it took when / if it finishes.

New issue with recent snaps:

Running the 20150224 (and *23) snapshot produces the following on Win
XP (yes, I know) if cygserver is running:

$ /usr/sbin/syslog-ng -F
      1 [main] syslog-ng 5776 C:\cygwin\usr\sbin\syslog-ng.exe: *** fatal error - Fetching account info from cygserver with wrong arg.type 2

.... and even if cygserver is not running:

$ getent group `id -G`
Domain Users:S-1-5-21-1643024071-179607362-792003330-513:10513:
      1 [main] getent 4120 C:\cygwin\bin\getent.exe: *** fatal error - Fetching account info from cygserver with wrong arg.type 2
Hangup

The 1.7.35-0.3 cygwin dll didn't trigger those symptoms.

tcsh still taking a long time here (also XP - 3+ minutes with
cygserver running).  bash & dash much quicker.  Maybe something with
the tcsh hash table (tcsh -f is fast)?  Doing 'env PATH=/bin tcsh'
helps some.  strace shows lots of mount_info::conv_to_posix_path
calls when PATH is not trimmed.

Other times, I've seen stalls with 'db' in nsswitch.conf and no
cygserver like so every time a new child tcsh.exe is forked:

00:00:00 [main] tcsh 4388 child_info::sync: n 2, waiting for subproc_ready(0x6BC) and child process(0x680)
00:05:00 [main] tcsh 4388 child_info::sync: wait failed, pid 2556, Win32 error 0

Why it's forking a number of child tcsh's is not clear even after
looking through the .csh files in /etc/profile.d Ahh... it must be
complete.tcsh.  Indeed - disabling that (by rename) helps a lot - lots
of back-tick commands in there.

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple