AWS server, m4.xlarge. Ubuntu 16.04 launched from AMI ami-2581aa40.
100gb EBS as root volume with 5000 provisioned iops.

Unfortunately it looks like ExeMgr crashed when you executed that query. Can you please turn on core files and send us the core file generated?

You can turn on core files by following these instructions:

https://mariadb.com/kb/en/library/system-troubleshooting-mariadb-columnstore/#enabledisable-core-files

If you wish to upload using our FTP server the details are:

ftp://ftp.mariadb.com/uploads

Thanks for the help. I've generated the core file and attached it here. File is named MCOL1182-core.ExeMgr.18965.gz

Many thanks. We will analyse this as soon as possible.

Hello, any update?

Hello, thank you for confirming the bug. I see that its due to be fixed in 1.1.4. Do you know when that is targeted for release?

Hi,
Yes, sorry. I diagnosed the core file when uploaded but hadn't finished the triage. The core file helped figure out where the crash is happening and it is reproducible.

We are releasing 1.1.3 first before we schedule 1.1.4. Unfortunately it is too late to fix this for 1.1.3. We typically do a 1.1 release every 1-2 months.

Ok, understood. Is there any chance that a fix would be backported to 1.1.2. Or is there a simple workaround that does not involve converting all tables to the columnstore engine?

Sorry, completely missed that last comment until now.

The last digit in out releases is an incremental bugfix version, we would not re-release a bugfix version. We instead release incremental bugfix releases. We may backport to 1.0 if there is a need for it.

I'm not yet aware of a workaround beyond the one you mentioned.

Could this be a possible workaround until fix?

select k.value as make, count
from study_response_1 s
left join keymap k on (k.key1 = s.col25 and k.keymap_group_id = (case when k.key1 = s.col25 then 1 else 0 end))
group by k.value;

Issue appears to be doOuterJoinOnFilter() executes "delete c" on the filter when making it a nullTree (which doesn't appear to be needed) and the filter is then used for the cross engine join.

For QA:

CREATE TABLE `keymap` (

  `keymap_id` int(11) NOT NULL,

  `keymap_group_id` int(11) NOT NULL,

  `key1` varchar(100) NOT NULL,

  `value` varchar(100) DEFAULT NULL,

  PRIMARY KEY (`keymap_id`),

  KEY `Refkeymap_group29` (`keymap_group_id`)

) ENGINE=InnoDB;

CREATE TABLE `study_response_1` (

  `record_id` int(11) NOT NULL,

  `col1` varchar(1000) DEFAULT NULL,

  `col2` varchar(1000) DEFAULT NULL,

  `col3` varchar(1000) DEFAULT NULL,

  `col4` varchar(1000) DEFAULT NULL

) ENGINE=Columnstore;

select k.value as make from study_response_1 s  left join keymap k on k.key1 = s.col2 and k.keymap_group_id = 5;

Will crash ExeMgr before patch, empty result after.

Fix not perfect yet apparently...

New patch is much more stable

Build 1.1.4-1 source

/root/columnstore/mariadb-columnstore-server
commit 5199dd1a096fd3457e8fc0508bf5fb24cedec435
Merge: fce3c5e e554e04
Author: David.Hall <david.hall@mariadb.com>
Date: Wed Apr 11 11:04:46 2018 -0500

Merge pull request #108 from mariadb-corporation/MCOL-1331

MCOL-1331 Fix CASE1.DM.sql

/root/columnstore/mariadb-columnstore-server/mariadb-columnstore-engine
[root@localhost mariadb-columnstore-engine]# git show
commit ae04b8a6877c87f3ed3566f2bf721bf285ca625f
Merge: 2ab632c dbcbd6c
Author: david hill <david.hill@mariadb.com>
Date: Tue Apr 10 10:55:56 2018 -0400

Merge pull request #438 from mariadb-corporation/MCOL-1323

MCOL-1323 cpimport Splitter has incorrect SIGPIPE mapping

diff --cc writeengine/splitter/we_splitterapp.cpp
index f52f362,0077ebd..402d2b0
mode 100755,100644..100755
— a/writeengine/splitter/we_splitterapp.cpp
+++ b/writeengine/splitter/we_splitterapp.cpp

Reproduced the issue in 1.1.2-1 with only one try
Reproduced the issue in 1.1.3-1 after few tries
Verified the fix in 1.1.4-1 by repeating the same query.