Hi, I have a table of Endpoints which looks like this: class Endpoint(db_models.Model): ... op = db_models.ForeignKey(ReportOp, on_delete=db_models.CASCADE) client = db_models.ForeignKey(Client, on_delete=db_models.CASCADE, null=True, blank=True) client_private = db_models.IntegerF…

Is it possible that the presence of the two transaction.atomic() calls is responsible for keep the Exclusive lock in place for the duration of the bigger operation? The answer to this is: absolutely yes. When you start nesting transaction.atomic blocks you are not creating more transactions at D…

Hi! How are you using this method? The returned queryset must be used within a transaction where you’ll need to handle the various race conditions. The docs try to point this out here: QuerySet API reference | Django documentation | Django

@CodenameTim Thanks for the reply. The caller looks like this: qs = row.locked_instance() with transaction.atomic(): for db_row in qs.all(): ...moderate amount of compute bound stuff... db_row.save() I believe that is what the docs call f…

How can there be a deadlock between two separate rows? The logs for deadlocks may seems strange at first sight, because it is logged the moment in which the deadlock occurs. What your logs says is: “process 15576 would like to lock id=2 (and is locking id=1) while 16953 would like to lock id=1 (a…

Do you have logs that show both of the locks?

@sevdog ok, but the locked queryset is defined to only contain a single row .filter(id=instance.id), so I don’t see why ordering would make a difference in my case? @CodenameTim I have the client side error log as well as the Postgres server side error log posted above. But that seemed less use tha…

@ShaheedHaque from the snippet you have posted you are right, it should not cause deadlock. Is there any outer transaction or a loop in which your code invoke the above snippets? Do you use ATOMIC_REQUESTS ?

[image] ShaheedHaque: ERROR: deadlock detected DETAIL: Process 15576 waits for ShareLock on transaction 31053599; blocked by process 16953. Process 16953 waits for ShareLock on transaction 31053597; blocked by process 15576. Process 15576: SELECT “paiyroll_endpoint”.“id”, “paiyroll_endpoint”.…

Thanks for all the replies… @sevdog No, there is no outer transaction. You also asked if ATOMIC_REQUESTS was set and I was about to say “yes” but then I read up on ATOMIC_REQUESTS, and it turns out I do not have it set, but this code is typically executed from a Celery context as opposed to a view …

Perhaps I need to ask the Postgres folk?

[image] ShaheedHaque: ERROR: deadlock detected DETAIL: Process 15576 waits for ShareLock on transaction 31053599; blocked by process 16953. Process 16953 waits for ShareLock on transaction 31053597; blocked by process 15576. Process 15576: SELECT “paiyroll_endpoint”.“id”, “paiyroll_endpoint”.…

Unexpected deadlock across two separate rows, using Postgres 17 and select_for_update().

Using Django Using the ORM

ShaheedHaque May 12, 2026, 6:08pm 16

@sevdog Thanks for confirming. I guess I need to rethink how this bit works…

Are ORM transaction.atomic()s tied to a thread?

Topic		Replies	Views
multiple atomic transactions Using the ORM	3	1160	December 4, 2023
prevent double delivery of unhandled database rows Using Django	5	2224	May 15, 2021
Operability during long running loads Using the ORM	15	393	February 6, 2025
Do transaction.atomic() locks DB? Using Django	2	5173	May 19, 2020
Does the select_for_update method work with the update method? Using the ORM	2	10702	September 14, 2023

Unexpected deadlock across two separate rows, using Postgres 17 and select_for_update().

Related topics