Segmentation fault restoring data to Azure VM

Jeffrey Durrence jeffrey.durrence at mcleanengineering.com
Tue Apr 23 12:28:26 PDT 2024


Regina,

As you suggested, I dropped the postgis_raster extension, exported, and the imported.

No errors!

Hopefully a record of this will be helpful to others who find themselves in this situation.

Thanks again for your input!


Jeffrey Durrence

McLean Engineering Company

www.mcleanengineering.com<http://www.mcleanengineering.com/>

Office: 229-985-1148

Mobile: 229-798-0480



________________________________
From: Regina Obe <lr at pcorp.us>
Sent: Tuesday, April 23, 2024 2:05 PM
To: Jeffrey Durrence <jeffrey.durrence at mcleanengineering.com>; postgis-users at lists.osgeo.org <postgis-users at lists.osgeo.org>
Subject: RE: Segmentation fault restoring data to Azure VM


Yah dropping the raster extension if you are not using it would be my first suggestion.



If you do



DROP EXTENSION postgis_raster;



And it successfully drops, then you definitely don’t have data tied to it so safe to drop before you do your pg_dump and restore.



I’ve seen people complain about random crashes with for example a bad netcdf driver that gets loaded when gdal lib is loaded.  So if you are running into one of those types of issues, not installing the raster extension will save you from that pain.



I do see you have Azure postgis compiled with GEOS 3.10.2, but you are running with 3.11.1.  In theory that shouldn’t be an issue since 3.10.2 is upward compatible with 3.11.



Everything else at a glance looks to be in shape.



Hope that helps,

Regina



From: Jeffrey Durrence <jeffrey.durrence at mcleanengineering.com>
Sent: Tuesday, April 23, 2024 1:21 PM
To: Regina Obe <lr at pcorp.us>; postgis-users at lists.osgeo.org
Subject: Re: Segmentation fault restoring data to Azure VM



Regina,



Thank you so much for your quick response (and for your contributions to the rest of us in your writings)!

On the physical server I have:
 POSTGIS="3.4.2 c19ce56" [EXTENSION] PGSQL="150" GEOS="3.10.2-CAPI-1.16.0" PROJ="8.2.1 NETWORK_ENABLED=OFF URL_ENDPOINT=https://cdn.proj.org USER_WRITABLE_DIRECTORY=/tmp/proj DATABASE_PATH=/usr/share/proj/proj.db" GDAL="GDAL 3.4.3, released 2022/04/22" LIBXML="2.9.13" LIBJSON="0.15" LIBPROTOBUF="1.3.3" WAGYU="0.5.0 (Internal)" (core procs from "3.4.1 ca035b9" need upgrade) RASTER (raster procs from "3.4.1 ca035b9" need upgrade)



On the Azure server I have
POSTGIS="3.4.2 c19ce56" [EXTENSION] PGSQL="150" GEOS="3.11.1-CAPI-1.17.1" (compiled against GEOS 3.10.2) PROJ="8.2.1 NETWORK_ENABLED=OFF URL_ENDPOINT=https://cdn.proj.org USER_WRITABLE_DIRECTORY=/tmp/proj DATABASE_PATH=/usr/share/proj/proj.db" GDAL="GDAL 3.4.3, released 2022/04/22" LIBXML="2.9.13" LIBJSON="0.15" LIBPROTOBUF="1.3.3" WAGYU="0.5.0 (Internal)" RASTER

There are some differences, though I'm not sure of the best strategy to reconcile this.  As for the raster extension, I don't use it for my workflows.  I could drop the postgis_raster extension and see if that allows my export/import to succeed without error.



-Jeffrey



Jeffrey Durrence

McLean Engineering Company

www.mcleanengineering.com<http://www.mcleanengineering.com/>

Office: 229-985-1148

Mobile: 229-798-0480



________________________________

From: Regina Obe <lr at pcorp.us<mailto:lr at pcorp.us>>
Sent: Tuesday, April 23, 2024 1:09 PM
To: Jeffrey Durrence <jeffrey.durrence at mcleanengineering.com<mailto:jeffrey.durrence at mcleanengineering.com>>; postgis-users at lists.osgeo.org<mailto:postgis-users at lists.osgeo.org> <postgis-users at lists.osgeo.org<mailto:postgis-users at lists.osgeo.org>>
Subject: RE: Segmentation fault restoring data to Azure VM



What does



SELECT postgis_full_version();



Run on your current setup vs. your Azure VM setup.



Might thinking is it might be a dependent library crashing.



Usually the issues are with postgis_raster extension since that drags in GDAL which drags in several more libraries.



Could also be an issue with PROJ.  As I recall at least on windows, I used to get crashes if PROJ_LIB path wasn’t set, but I think that issue has gone away with recent releases.



Not sure if it was ever an issue on Linux or Unix.



Hope that helps,

Regina



From: Jeffrey Durrence <jeffrey.durrence at mcleanengineering.com<mailto:jeffrey.durrence at mcleanengineering.com>>
Sent: Tuesday, April 23, 2024 11:42 AM
To: postgis-users at lists.osgeo.org<mailto:postgis-users at lists.osgeo.org>
Subject: Segmentation fault restoring data to Azure VM



Greetings,



I am migrating PostgreSQL/PostGIS databases from a physical server with these specs:
Host OS: Ubuntu 22.04.4 LTS x64 (32GB RAM, 8GB Swap, 1TB disk space, XEON E5-1660 CPU)
Postgres: 15.6.1-pgdg22.04+1
PostGIS: 3.4.2



To an Azure VM with

Host OS: Ubuntu 22.04.4 LTS x64 (32GB RAM, 8GB Swap, 512GB disk space, XEON 8171M) (again, this is in Azure VM)
Postgres: 15.6.1-pgdg22.04+1
PostGIS: 3.4.2





In the past, I have migrated the same data from on the same physical hardware for changes in host O/S, Postgres version, and/or PostGIS version. I've also migrated from one physical host to another.  When doing so, I try to make sure that my software packages are up to date and that the packages in the destination environment match.  I did this for this planned environment as well.  Because I know that some of my dbs have several years of history in them, I export these dbs individually using pg_dump (using the custom format when possible but in some cases I use the plain text option) and pg_restore on the destination host.



During the process of migrating data in this manner, I observed that restoration of many of my dbs failed.  I did lots of testing with this and have made these additional observations:

  1.  I am able to restore my export files to another physical host with the same software environment without error.  (the errors occur when importing to virtual environment).

  1.  If I split my export/import into separate steps for the schema and data, it is the schema import that produces the error, not the data import.

  1.  When the error occurs, the db process terminates with a segmentation vault (see log output below), and the db process restarts.

  1.  With logging of all statements enabled, I can see that the segmentation vault occurs at the absolute end of the import, when it seems that all commands are completed.  So long as I wait on the server process to recover, I have no errors when I import the data following the error.

  1.  The db imports that produce these errors seem to be those with more history (e.g., it's more likely that these were created with an older version of postGIS and subsequently upgraded to the versions they have now).

  1.  The errors do seem related to the installation of the PostGIS extension.  If I purposely drop the extension from the source database before I export, the resulting file can be imported in the target environment without error.  This is obviously undesirable as it results in my losing all of the geometry info from tables, views, etc. in the db.  I only did this to experiment.



My typical restore command looks like this:
pg_restore --verbose --format=custom --dbname=my_db /tmp/my_db_schema_only.backup

(plus some options to capture the output)



Here's a look at what the command line output looks like when this happens:

LOG:  server process (PID 52476) was terminated by signal 11: Segmentation fault

LOG:  terminating any other active server processes

LOG:  all server processes terminated; reinitializing

LOG:  database system was interrupted; last known up at 2024-04-23 02:16:49 UTC

LOG:  database system was not properly shut down; automatic recovery in progress

LOG:  redo starts at 5/51381960

WARNING:  could not open directory "base/696372": No such file or directory

CONTEXT:  WAL redo at 5/51381A48 for Database/DROP: dir 1663/696372

WARNING:  some useless files may be left behind in old database directory "base/696372"

CONTEXT:  WAL redo at 5/51381A48 for Database/DROP: dir 1663/696372

LOG:  invalid record length at 5/52679558: expected at least 24, got 0

LOG:  redo done at 5/526794E0 system usage: CPU: user: 0.07 s, system: 0.01 s, elapsed: 0.10 s

LOG:  checkpoint starting: end-of-recovery immediate wait

LOG:  checkpoint complete: wrote 2699 buffers (16.5%); 0 WAL file(s) added, 1 removed, 0 recycled;

B; lsn=5/52679558, redo lsn=5/52679558

LOG:  database system is ready to accept connections



In the log file, the section written as the error occurs is something like this...

[60419] [6627bcbb.ec03] [2024-04-23 13:50:54.813 UTC] [0]: LOG:  statement: GRANT ALL ON TABLE my_schema.my_table TO my_user_name;

[60419] [6627bcbb.ec03] [2024-04-23 13:50:54.815 UTC] [0]: LOG:  statement: -- Completed on 2024-04-23 13:50:54 UTC

[60419] [6627bcbb.ec03] [2024-04-23 13:50:54.815 UTC] [0]: LOG:  statement: --

        -- PostgreSQL database dump complete

        --

[59940] [6627b39f.ea24] [2024-04-23 13:50:54.943 UTC] [0]: LOG:  server process (PID 60419) was terminated by signal 11: Segmentation fault

[59940] [6627b39f.ea24] [2024-04-23 13:50:54.943 UTC] [0]: LOG:  terminating any other active server processes

[59940] [6627b39f.ea24] [2024-04-23 13:50:54.945 UTC] [0]: LOG:  all server processes terminated; reinitializing

[60423] [6627bcbf.ec07] [2024-04-23 13:50:55.007 UTC] [0]: LOG:  database system was interrupted; last known up at 2024-04-23 13:50:48 UTC

[60423] [6627bcbf.ec07] [2024-04-23 13:51:05.511 UTC] [0]: LOG:  syncing data directory (fsync), elapsed time: 10.00 s, current path: ./base/897556/1249

[60423] [6627bcbf.ec07] [2024-04-23 13:51:12.554 UTC] [0]: LOG:  database system was not properly shut down; automatic recovery in progress

[60423] [6627bcbf.ec07] [2024-04-23 13:51:12.606 UTC] [0]: LOG:  redo starts at E/3A8785A8

[60423] [6627bcbf.ec07] [2024-04-23 13:51:12.607 UTC] [0]: WARNING:  could not open directory "base/562219": No such file or directory

[60423] [6627bcbf.ec07] [2024-04-23 13:51:12.607 UTC] [0]: CONTEXT:  WAL redo at E/3A878690 for Database/DROP: dir 1663/562219

[60423] [6627bcbf.ec07] [2024-04-23 13:51:12.607 UTC] [0]: WARNING:  some useless files may be left behind in old database directory "base/562219"

[60423] [6627bcbf.ec07] [2024-04-23 13:51:12.607 UTC] [0]: CONTEXT:  WAL redo at E/3A878690 for Database/DROP: dir 1663/562219

[60423] [6627bcbf.ec07] [2024-04-23 13:51:12.704 UTC] [0]: LOG:  invalid record length at E/3BACDAF8: expected at least 24, got 0

[60423] [6627bcbf.ec07] [2024-04-23 13:51:12.704 UTC] [0]: LOG:  redo done at E/3BACDA80 system usage: CPU: user: 0.07 s, system: 0.01 s, elapsed: 0.09 s

[60424] [6627bcbf.ec08] [2024-04-23 13:51:12.804 UTC] [0]: LOG:  checkpoint starting: end-of-recovery immediate wait

[60424] [6627bcbf.ec08] [2024-04-23 13:51:13.143 UTC] [0]: LOG:  checkpoint complete: wrote 2592 buffers (15.8%); 0 WAL file(s) added, 1 removed, 0 recycled; writ

9 s, average=0.001 s; distance=18773 kB, estimate=18773 kB; lsn=E/3BACDAF8, redo lsn=E/3BACDAF8

[59940] [6627b39f.ea24] [2024-04-23 13:51:13.156 UTC] [0]: LOG:  database system is ready to accept connections





I searched the mail archive and performed some additional web searches based on my observations.  Unfortunately, I haven't found a similar-enough incident where I might get suggestions for further troubleshooting.  In over 10 years of using PostgreSQL+PostGIS, I've never had to dig this deeply to determine the source of error.  I'm looking for advice for further troubleshooting or for shared experiences in similar situations



Thank you for your time!



Jeffrey Durrence

McLean Engineering Company

www.mcleanengineering.com<http://www.mcleanengineering.com/>

Office: 229-985-1148

Mobile: 229-798-0480


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/postgis-users/attachments/20240423/a1c9115e/attachment.htm>


More information about the postgis-users mailing list