[Spread-users] GGT bug
Ryan Caudy
caudy at jhu.edu
Mon Aug 26 20:37:34 EDT 2002
Yair Amir wrote:
>Hi,
>
>Theo> On Monday, August 26, 2002, at 06:29 , Ryan Caudy wrote:
>
>
>>>Currently there exists a bug in the groups state transfer that will
>>>cause spread to hang under certain conditions caused by cascading
>>>network changes. The fix will be included in the next release of
>>>Spread. However, for those who need/want a quick fix now, the
>>>following *minor* code change should solve the problem.
>>>
>>>The affected file is groups.c, which should be under spread/daemon.
>>>
>>>Line 1780 currently reads as:
>>> "memb_id_ptr = &buf[ Message_get_data_header_size() ];"
>>>
>>>To solve the problem, change this to:
>>> "memb_id_ptr = buf;"
>>>
>>>
>
>Theo> If this bug is tickled, what state are the daemons left in? state=1
>Theo> gstate=1 or something more obviously broken?
>
>state= OP gstate= GGT. They will not complete the state transfer.
>
>This bug was introduced when we merged the objects subsystem.
>
The error is in the GGT code, but the bug causes the daemon to stay in
GGATHER, and will never allow the state transfer to complete... network
conditions could cause a shift back to GGT, but the state should return
to GGATHER after the daemon membership completes. So, you should expect
to see state=1 (OP), gstate=3 (GGATHER) for a long time if this bug is
triggered.
--Ryan
More information about the Spread-users
mailing list