Date: 2011-09-19 22:21:27 +0200
From: @grobian
To: SQL devs <>
Version: 11.5.1 (Aug2011) [obsolete]

Last updated: 2011-09-30 10:58:46 +0200

Comment 16305

Date: 2011-09-19 22:21:27 +0200
From: @grobian

Testing stops with defunct processes (mserver5?) repeatedly in Aug2011 and default branches on all combinations on sparc (s32, s48, s64) for test/null/null_in_case.sql. As a result, everything hangs for eternity, since Python apparently doesn't receive/handle the signal its child crashed/cored.

As a result, entire testing blocks until the appropriate preap or kills of Mtest.py are done on the build sun.

Comment 16306

Date: 2011-09-20 09:45:46 +0200
From: @sjoerdmullender

I think the null_in_case test is probably a red herring.
The very first sql test already ends in a timeout, as do many other tests (at least on GNU-Solaris-sparc).

Comment 16310

Date: 2011-09-20 11:15:20 +0200
From: @grobian

Very well possible. However, after hours (and even days) of waiting, all testing runs always hang at this particular test.

Comment 16312

Date: 2011-09-20 21:22:50 +0200
From: @sjoerdmullender

Changeset 9d839b4fd018 made by Sjoerd Mullender sjoerd@acm.org in the MonetDB repo, refers to this bug.

For complete details, see http//devmonetdborg/hg/MonetDB?cmd=changeset;node=9d839b4fd018

Changeset description:

When clearing variables also clear eolife field.
Not clearing the field could cause a bogus dependency in the dataflow
scheduler which in turn could cause a circular dependecy among
instructions.
This fixes bug #2865 and the server timeouts of bug #2888.
(The problem that Mtest.py has with all the hanging servers is not
solved but may not be a problem anymore.)

Comment 16314

Date: 2011-09-21 11:08:39 +0200
From: @sjoerdmullender

Changeset 02c039857104 made by Sjoerd Mullender sjoerd@acm.org in the MonetDB repo, refers to this bug.

For complete details, see http//devmonetdborg/hg/MonetDB?cmd=changeset;node=02c039857104

Changeset description:

Add assertions about inter-instruction dependencies.
Instructions should only depend on earlier instructions.
These asserts would probably have made finding the problem with bugs
2865 and 2888 a lot easier...

Comment 16353

Date: 2011-09-30 10:58:46 +0200
From: @grobian

Released in Aug2011-SP1