The `user_activity_interval` worker calls:
```python3
last = UserActivityInterval.objects.filter(user_profile=user_profile).order_by("-end")[0]
`````
Which results in a query like:
```sql
SELECT "zerver_useractivityinterval"."id", "zerver_useractivityinterval"."user_profile_id", "zerver_useractivityinterval"."start", "zerver_useractivityinterval"."end" FROM "zerver_useractivityinterval" WHERE "zerver_useractivityinterval"."user_profile_id" = 12345 ORDER BY "zerver_useractivityinterval"."end" DESC LIMIT 1
```
For users which have at least one matching row, this results in a
query plan like:
```
Limit (cost=0.56..711.38 rows=1 width=24) (actual time=0.078..0.078 rows=1 loops=1)
-> Index Scan Backward using zerver_useractivityinterval_7f021a14 on zerver_useractivityinterval (cost=0.56..1031399.46 rows=1451 width=24) (actual time=0.077..0.078 rows=1 loops=1)
Filter: (user_profile_id = 12345)
Rows Removed by Filter: 98
Planning Time: 0.059 ms
Execution Time: 0.088 ms
```
But for users that have just been created, with no matching rows, this
is considerably more expensive:
```
Limit (cost=0.56..711.38 rows=1 width=24) (actual time=10798.146..10798.146 rows=0 loops=1)
-> Index Scan Backward using zerver_useractivityinterval_7f021a14 on zerver_useractivityinterval (cost=0.56..1031399.46 rows=1451 width=24) (actual time=10798.145..10798.145 rows=0 loops=1)
Filter: (user_profile_id = 12345)
Rows Removed by Filter: (count of every single row in the table, redacted)
Planning Time: 0.053 ms
Execution Time: 10798.158 ms
```
Regular vacuuming can force the use of the index on `user_profile_id`
as long as there are few enough users, which is fast -- however, at
some point, the query planner decides that is insufficiently specific,
always chooses the effective-whole-table-scan.
Add an index on `(user_profile_id, end)`, which is expected to be
sufficiently specific that it is used even with large numbers of user
profiles.
Ref #19250.
Alex Vandiver