Setup
=====

Testcode: https://git.samba.org/?p=slow/samba.git;a=shortlog;h=refs/heads/write-time
SMB1 test: "base.delaywrite.delayed update of write time loop" (hardcoded delay of 0)
SMB2 test: smb2.delaywrite.delayed-write2 --option=torture:delay_ms=X

Servers:
* Windows 2003 R2:	172.18.103.94
* Windows 7:   		172.18.103.129
* Windows 2008 R1:	172.18.103.81
* Windows 2016:		172.18.103.204
* Windows 2019:		192.168.122.99

SMB1
====

Windows 2003 R2
---------------

* one delayed mtime update
* no subsequent updates
* update on close when delayed update pending

Initial write time Wed Oct  9 13:31:37 2019 CEST
Server updated write_time after 4.65 seconds (correct)

Windows 7
---------

* delayed update every 2 s on avg, can be only 0.8 s and up to 3 s
* update on close when delayed update pending

Initial write time Wed Oct  9 13:35:31 2019 CEST
Server updated write_time after 2.97 seconds (correct)
Server updated write_time after 1.57 seconds (correct)
Server updated write_time after 2.01 seconds (correct)
Server updated write_time after 2.00 seconds (correct)
Server updated write_time after 2.00 seconds (correct)
Server updated write_time after 1.85 seconds (correct)
Server updated write_time after 1.99 seconds (correct)
Server updated write_time after 2.00 seconds (correct)
Server updated write_time after 0.88 seconds (correct)
Server updated write_time after 2.00 seconds (correct)

Windows 2016
------------

* delayed update every 1.3 s on avg
* update on close when delayed update pending

Initial write time Wed Oct  9 14:18:58 2019 CEST
Server updated write_time after 0.21 seconds (correct)
Server updated write_time after 1.48 seconds (correct)
Server updated write_time after 1.32 seconds (correct)
Server updated write_time after 1.32 seconds (correct)
Server updated write_time after 1.30 seconds (correct)
Server updated write_time after 1.30 seconds (correct)
Server updated write_time after 1.30 seconds (correct)
Server updated write_time after 1.32 seconds (correct)
Server updated write_time after 1.32 seconds (correct)
Server updated write_time after 1.30 seconds (correct)

SMB2
====

Windows 7
---------

--option=torture:delay_ms=0

Using delay of 0 ms
Server updated write_time after 0.605584 s
Server updated write_time after 1.967016 s
Server updated write_time after 2.118526 s
Server updated write_time after 1.966535 s
Server updated write_time after 1.966680 s
Server updated write_time after 1.967038 s
Server updated write_time after 1.966066 s
Server updated write_time after 2.117810 s
Server updated write_time after 1.966421 s
Server updated write_time after 1.966136 s

Windows 2008 R1
---------------

--option=torture:delay_ms=0

Using delay of 0 ms
Server updated write_time after 0.908788 s
Server updated write_time after 1.815782 s
Server updated write_time after 1.967364 s
Server updated write_time after 2.119486 s
Server updated write_time after 1.967551 s
Server updated write_time after 1.967161 s
Server updated write_time after 1.967556 s
Server updated write_time after 1.072425 s
Server updated write_time after 1.968118 s
Server updated write_time after 1.966277 s

Windows 2016
------------

* strange behaviour, depends on time between create/IO and getinfo
  threshold is between 900-1000ms for the delay

* below the threshold the mtime isn't updated when queried with SMB2 GETINFO,
  only the final close updates mtime

* above the threshold write IO always triggers an delayed update. The first
  mtime query after the write IO is done after 900 ms, but the mtime gets
  updated a bit later, ~1.3 s after the write IO

--option=torture:delay_ms=900

Using delay of 900 ms
No update within 10 s
No update within 10 s
No update within 10 s
No update within 10 s
No update within 10 s
No update within 10 s
No update within 10 s
No update within 10 s
No update within 10 s
No update within 10 s

--option=torture:delay_ms=1000

Using delay of 1000 ms
Server updated write_time after 2.060728 s
Server updated write_time after 1.304737 s
Server updated write_time after 1.304404 s
Server updated write_time after 1.304762 s
Server updated write_time after 1.456069 s
Server updated write_time after 1.455422 s
Server updated write_time after 1.304588 s
Server updated write_time after 1.455730 s
Server updated write_time after 1.304448 s
Server updated write_time after 1.455462 s

Windows 2019
------------

* strange behaviour: depends on time between write IO and querying mtime.
  Threshold is between 0 ms and 15 ms for the time between write IO and querying mtime

* below the threshold the mtime is updated immediately after every second write IO

* if the last IO didn't trigger an mtime update, then also the close doesn't
  update it (!)

* above the threshold every write IO triggers immediate mtime update

* bottom line: querying mtime too early after write IO prevents mtime
  updates. This applies to every second write IO.

--option=torture:delay_ms=0

Using delay of 0 ms
No update within 10 s
Server updated write_time after 0.000638 s
No update within 10 s
Server updated write_time after 0.000496 s
No update within 10 s
Server updated write_time after 0.000559 s
No update within 10 s
Server updated write_time after 0.000627 s
No update within 10 s
Server updated write_time after 0.000782 s
No update within 10 s

--option=torture:delay_ms=15

Using delay of 15 ms
Server updated write_time after 0.015542 s
Server updated write_time after 0.016005 s
Server updated write_time after 0.015416 s
Server updated write_time after 0.016130 s
Server updated write_time after 0.015710 s
Server updated write_time after 0.016943 s
Server updated write_time after 0.015579 s
Server updated write_time after 0.016992 s
Server updated write_time after 0.015489 s
Server updated write_time after 0.016931 s

Samba SMB 1+2
=============

* one delayed mtime update
* no subsequent updates
* update on close when delayed update pending
* same as Windows 2003 R2 behaviour