Orientdb: All the metadata of indexes are lost after automatic index rebuilds

[rdelangh]

OrientDB Version, operating system, or hardware.

• v2.2.14

Operating System

• [x] Linux

Expected behavior and actual behavior

All the metadata of indexes are lost after automatic index rebuilds...

We set the metadata

allowLeadingWildcard":true

for all of our Lucene indexes.

After a restart of our server (we had a sudden death of the system), a rebuild was started of nearly all of the indexes, including the Lucene indexes.
Now, I notice that such rebuilt Lucene indexes have not these metadata anymore. Obviously resulting in errors for our queries where we use such wildcards.

[robfrank]

Hi, even if the clause allowLeadingWildcard":true involves only the Query parser behaviour, it is necessary to drop and rebuild the index to re-enable it.
I'm afraid of that. I've got other bugs in the backlog, I'll try to do my best to solve it.
Thanks for reporting

[robfrank]

Hi, even if the clause allowLeadingWildcard":true involves only the Query parser behaviour, it is necessary to drop and rebuild the index to re-enable it.
I'm afraid of that. I've got other bugs in the backlog, I'll try to do my best to solve it.
Thanks for reporting

[robfrank]

Hi, even if the clause allowLeadingWildcard":true involves only the Query parser behaviour, it is necessary to drop and rebuild the index to re-enable it.
I'm afraid of that. I've got other bugs in the backlog, I'll try to do my best to solve it.
Thanks for reporting

[rdelangh]

@robfrank

ok, thx for the confirmation, at least we know now it is a known bug. Which will be resolved very soon...;-)

Meanwhile, dropping and recreating these indexes. This lasts for about 1.5hrs per index (20-30M docs per class, multiple indexes per class). Only 17 indexes to go... :-(

[robfrank]

I was not clear. We didn't have knowledge of the bug, but if after a crash the index rebuild procedure doesn't load the metadata, well, the index needs to be rebuilt. So, I'm trying to reproduce the bug, that's not so easy (hammering a running server every time seems to be too much expensive :) )

[rdelangh]

hammering a running server every time seems to be too much expensive

no big deal: one possibility is to launch a virtual server, in that you run ODB server and get it busy; then from the host machine, abruptly stop the virtual server and disable abruptly its access to the disk device. Done.

[robfrank]

I've managed to replicate the bug.
Good: the index is recreated using the right metadata: indexes are rebuilt right!.
Bad: sometimes, after rebuild, the index is re-opened with wrong metadata.
Note that it happens only sometimes (random).
Possible workaround: after rebuild, switch off and the on the server.
We are working to find a fix.

[rdelangh]

@robfrank thx!
cool, that's already a big step in finding a fix.

[robfrank]

FYI, @taburet is working actively on it.

[robfrank]

We are doing final tests, the fix is on SNAPSHOTs and will be part of 2.2.17

[robfrank]

Our tests are always green. I close and move on 2.2.17 milestone.