[Issue 10316] New: build/lib/mdb.c:5882: Assertion 'IS_BRANCH(mc->mc_pg[mc->mc_top])' failed in mdb_cursor_sibling()
https://bugs.openldap.org/show_bug.cgi?id=10316
Issue ID: 10316 Summary: build/lib/mdb.c:5882: Assertion 'IS_BRANCH(mc->mc_pg[mc->mc_top])' failed in mdb_cursor_sibling() Product: LMDB Version: unspecified Hardware: All OS: All Status: UNCONFIRMED Keywords: needs_review Severity: normal Priority: --- Component: liblmdb Assignee: bugs@openldap.org Reporter: stefan@konink.de Target Milestone: ---
I am developing ETL software in Python and I am using lmdb as backend. For this I am using https://github.com/jnwatson/py-lmdb
In this big process I am resizing the map_size dynamically. I would say it is certainly possible that I am reading at the same time. I end up in the following error:
build/lib/mdb.c:5882: Assertion 'IS_BRANCH(mc->mc_pg[mc->mc_top])' failed in mdb_cursor_sibling()
I ran my application again with a predetermined map size, and it did not assert. Since it is a single shot, I don't know if I can be certain that it therefore does not happen.
When searching for it I found the following issue from 2017, suggesting it can be reproduced when running the performance test.
https://github.com/bnclabs/dbperf/issues/1
https://bugs.openldap.org/show_bug.cgi?id=10316
--- Comment #1 from Howard Chu hyc@openldap.org --- Since you're only seeing this in code that performs a resize, it sounds more like you haven't correctly prevented other threads from using the DB while the resize is occurring. That's not an LMDB issue.
https://bugs.openldap.org/show_bug.cgi?id=10316
--- Comment #2 from Stefan de Konink stefan@konink.de --- Just to confirm. LMDB can do reading in threads and other processes while one other thread or process is writing. But for resizing it must halt all read operations?
This sounds a lot less problematic than the issue from 2017 :)
https://bugs.openldap.org/show_bug.cgi?id=10316
Howard Chu hyc@openldap.org changed:
What |Removed |Added ---------------------------------------------------------------------------- Status|UNCONFIRMED |RESOLVED Resolution|--- |INVALID Keywords|needs_review |
--- Comment #3 from Howard Chu hyc@openldap.org --- (In reply to Stefan de Konink from comment #2)
Just to confirm. LMDB can do reading in threads and other processes while one other thread or process is writing. But for resizing it must halt all read operations?
That is exactly what the documentation says.
http://www.lmdb.tech/doc/group__mdb.html#gaa2506ec8dab3d969b0e609cd82e619e5
https://bugs.openldap.org/show_bug.cgi?id=10316
--- Comment #4 from Stefan de Konink stefan@konink.de --- Does this also mean that mdb_dbi_open must be closed prior to resizing? Because I currently retain these too. I haven't had issues with that. And quite naively used it.
Is the best practice then to:
1. wait for cursors to finish 2. close all open dbs 3. close the current env 4. resize env 5. open env
https://bugs.openldap.org/show_bug.cgi?id=10316
--- Comment #5 from Howard Chu hyc@openldap.org --- (In reply to Stefan de Konink from comment #4)
Does this also mean that mdb_dbi_open must be closed prior to resizing? Because I currently retain these too. I haven't had issues with that. And quite naively used it.
Is the best practice then to:
- wait for cursors to finish
- close all open dbs
- close the current env
- resize env
- open env
That is not what the doc says.
Read it more carefully. http://www.lmdb.tech/doc/group__mdb.html#gaa2506ec8dab3d969b0e609cd82e619e5
There's no LMDB bug here, this ticket is closed.
https://bugs.openldap.org/show_bug.cgi?id=10316
Quanah Gibson-Mount quanah@openldap.org changed:
What |Removed |Added ---------------------------------------------------------------------------- Status|RESOLVED |VERIFIED
-
openldap-its＠openldap.org