#24554: Migrations taking up to an hour to run
-------------------------------------+-------------------------------------
Reporter: ryanahall | Owner: nobody
Type: | Status: new
Cleanup/optimization |
Component: Migrations | Version: 1.7
Severity: Normal | Resolution:
Keywords: | Triage Stage:
| Unreviewed
Has patch: 0 | Needs documentation: 0
Needs tests: 0 | Patch needs improvement: 0
Easy pickings: 0 | UI/UX: 0
-------------------------------------+-------------------------------------
Description changed by ryanahall:
Old description:
> We're currently running django 1.7.2 and been having performance issues
> with migrate/makemigrations for the last few months now, but until
> recently it hasn't caused us too much trouble. Recently,
> migrate/makemigrations calls have reached a point where they take up to
> 45minutes to run. Unfortunately, this is due to the fact that we have
> around 700 models defined with almost 850 migrations since the initial
> project. Regardless, I've been looking for any ways to speed this process
> up in 1.7.2/1.7.7/1.8b2/master and found that we have a significant
> bottleneck with the way states are cloned.
>
> Out of curiousity, I profiled latest master against a set of ~7
> migrations to run and found this:
> {{{
> ncalls tottime percall cumtime percall filename:lineno(function)
> 1 0.001 0.001 2687.730 2687.730 manage.py:2(<module>)
> 1 0.000 0.000 2687.669 2687.669
> django/core/management/__init__.py:325(execute_from_command_line)
> 1 0.000 0.000 2687.669 2687.669
> django/core/management/__init__.py:265(execute)
> 1 0.000 0.000 2680.113 2680.113
> django/core/management/base.py:326(run_from_argv)
> 1 0.002 0.002 2680.111 2680.111
> django/core/management/base.py:361(execute)
> 1 0.198 0.198 2678.912 2678.912
> django/core/management/commands/migrate.py:50(handle)
> 1 10.405 10.405 2676.084 2676.084
> django/db/migrations/executor.py:65(migrate)
> 803 0.024 0.000 2581.336 3.215
> django/db/migrations/migration.py:72(mutate_state)
> 7528892 36.422 0.000 1578.058 0.000
> django/db/migrations/state.py:468(construct_fields)
> 4137 0.504 0.000 1470.027 0.355
> django/db/migrations/state.py:81(reload_model)
> 847 0.012 0.000 1192.585 1.408
> django/db/migrations/state.py:135(clone)
> 847 0.939 0.001 1171.224 1.383
> django/db/migrations/state.py:138(<dictcomp>)
> 402948 6.238 0.000 1170.285 0.003
> django/db/migrations/state.py:488(clone)
> }}}
>
> In migration.py(mutate_state), the project state gets cloned before
> applying the operations everytime. I realize that the project state is
> intended to be immutable since the intermediary states need to be used
> most of the time, but I'm curious if it is required. During phase 1 of
> migration execution, the state gets cloned on every iteration inside
> mutate_state at line 90:
> {{{#!python
> for migration, _ in full_plan:
> if migration in migrations_to_run:
> states[migration] = state.clone()
> state = migration.mutate_state(state) # state is cloned
> inside
> }}}
> and also in migrations/graph.py(make_state) at line 274:
> {{{#!python
> for node in plan:
> project_state = self.nodes[node].mutate_state(project_state)
> }}}
>
> If I modify mutate_state and add an option to not preserve (no clone) the
> passed in state in these specific cases, the migration time is cut in
> half:
> {{{
> ncalls tottime percall cumtime percall filename:lineno(function)
> 1 0.001 0.001 955.337 955.337 manage.py:2(<module>)
> 1 0.000 0.000 955.269 955.269
> django/core/management/__init__.py:325(execute_from_command_line)
> 1 0.000 0.000 955.269 955.269
> django/core/management/__init__.py:265(execute)
> 1 0.000 0.000 946.453 946.453
> django/core/management/base.py:326(run_from_argv)
> 1 0.002 0.002 946.450 946.450
> django/core/management/base.py:361(execute)
> 1 0.169 0.169 945.012 945.012
> django/core/management/commands/migrate.py:50(handle)
> 1 0.006 0.006 941.582 941.582
> django/db/migrations/executor.py:65(migrate)
> 4137 0.387 0.000 869.433 0.210
> django/db/migrations/state.py:81(reload_model)
> 803 0.020 0.000 863.434 1.075
> django/db/migrations/migration.py:72(mutate_state)
> 100499 28.233 0.000 405.016 0.004
> django/apps/registry.py:323(clear_cache)
> 1753 0.027 0.000 399.287 0.228
> django/db/migrations/operations/fields.py:43(state_forwards)
> }}}
>
> My question is: does the state need to be cloned in mutate_state during
> phase 1? The intermediary states are thrown away besides the explicit
> calls to clone here:
> {{{#!python
> if migration in migrations_to_run:
> states[migration] = state.clone()
> }}}
>
> I can create create a pull request if you'd like to see.
New description:
We're currently running django 1.7.2 and been having performance issues
with migrate/makemigrations for the last few months now, but until
recently it hasn't caused us too much trouble. Recently,
migrate/makemigrations calls have reached a point where they take up to
45minutes to run. Unfortunately, this is due to the fact that we have
around 700 models defined with almost 850 migrations since the initial
project. Regardless, I've been looking for any ways to speed this process
up in 1.7.2/1.7.7/1.8b2/master and found that we have a significant
bottleneck with the way states are cloned.
Out of curiousity, I profiled latest master against a set of ~7 migrations
to run and found this:
{{{
ncalls tottime percall cumtime percall filename:lineno(function)
1 0.001 0.001 2687.730 2687.730 manage.py:2(<module>)
1 0.000 0.000 2687.669 2687.669
django/core/management/__init__.py:325(execute_from_command_line)
1 0.000 0.000 2687.669 2687.669
django/core/management/__init__.py:265(execute)
1 0.000 0.000 2680.113 2680.113
django/core/management/base.py:326(run_from_argv)
1 0.002 0.002 2680.111 2680.111
django/core/management/base.py:361(execute)
1 0.198 0.198 2678.912 2678.912
django/core/management/commands/migrate.py:50(handle)
1 10.405 10.405 2676.084 2676.084
django/db/migrations/executor.py:65(migrate)
803 0.024 0.000 2581.336 3.215
django/db/migrations/migration.py:72(mutate_state)
7528892 36.422 0.000 1578.058 0.000
django/db/migrations/state.py:468(construct_fields)
4137 0.504 0.000 1470.027 0.355
django/db/migrations/state.py:81(reload_model)
847 0.012 0.000 1192.585 1.408
django/db/migrations/state.py:135(clone)
847 0.939 0.001 1171.224 1.383
django/db/migrations/state.py:138(<dictcomp>)
402948 6.238 0.000 1170.285 0.003
django/db/migrations/state.py:488(clone)
}}}
In migration.py(mutate_state), the project state gets cloned before
applying the operations everytime. I realize that the project state is
intended to be immutable since the intermediary states need to be used
most of the time, but I'm curious if it is required. During phase 1 of
migration execution, the state gets cloned on every iteration inside
mutate_state at line 90:
{{{#!python
for migration, _ in full_plan:
if migration in migrations_to_run:
states[migration] = state.clone()
state = migration.mutate_state(state) # state is cloned
inside
}}}
and also in migrations/graph.py(make_state) at line 274:
{{{#!python
for node in plan:
project_state = self.nodes[node].mutate_state(project_state)
}}}
If I modify mutate_state and add an option to not preserve (no clone) the
passed in state in these specific cases, the migration time is cut in
half:
{{{
ncalls tottime percall cumtime percall filename:lineno(function)
1 0.001 0.001 955.337 955.337 manage.py:2(<module>)
1 0.000 0.000 955.269 955.269
django/core/management/__init__.py:325(execute_from_command_line)
1 0.000 0.000 955.269 955.269
django/core/management/__init__.py:265(execute)
1 0.000 0.000 946.453 946.453
django/core/management/base.py:326(run_from_argv)
1 0.002 0.002 946.450 946.450
django/core/management/base.py:361(execute)
1 0.169 0.169 945.012 945.012
django/core/management/commands/migrate.py:50(handle)
1 0.006 0.006 941.582 941.582
django/db/migrations/executor.py:65(migrate)
4137 0.387 0.000 869.433 0.210
django/db/migrations/state.py:81(reload_model)
803 0.020 0.000 863.434 1.075
django/db/migrations/migration.py:72(mutate_state)
100499 28.233 0.000 405.016 0.004
django/apps/registry.py:323(clear_cache)
1753 0.027 0.000 399.287 0.228
django/db/migrations/operations/fields.py:43(state_forwards)
}}}
My question is: does the state need to be cloned in mutate_state during
phase 1? The intermediary states are thrown away besides the explicit
calls to clone here:
{{{#!python
if migration in migrations_to_run:
states[migration] = state.clone()
}}}
Here's an example commit with this change:
https://github.com/ryanahall/django/commit/a4777e11d8950bd86b07d00ff2d291567c7ba0ff
--
--
Ticket URL: <https://code.djangoproject.com/ticket/24554#comment:2>
Django <https://code.djangoproject.com/>
The Web framework for perfectionists with deadlines.
--
You received this message because you are subscribed to the Google Groups
"Django updates" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To post to this group, send email to [email protected].
To view this discussion on the web visit
https://groups.google.com/d/msgid/django-updates/067.cfa225511abcac8d4a19af8994507763%40djangoproject.com.
For more options, visit https://groups.google.com/d/optout.
