Reduce the size of data for analyze tests. #6674
Conversation
| await con.query_single('select _set_seqscan("off")') | ||
| no_ex = '(execute := False) ' if not execute else '' | ||
| return json.loads(await (con or self.con).query_single( | ||
| return json.loads(await con.query_single( |
There was a problem hiding this comment.
Should we make it configurable whether we turn of seqscan? Maybe some of tests could leave it on which would allow us to reduce test churn. Maybe not really, though.
There was a problem hiding this comment.
For analyze tests leaving it on doesn't really help us, I think. If in the future we run into analyze tests that specifically need seqscan on then we can make it configurable.
There was a problem hiding this comment.
To be clear regarding the test churn, part of this is also dramatically reducing the number of tests objects created. So the results of the original tests would likely not align even if you turned the seqscan on for them.
|
On my machine the startup for just the analyze tests went from ~55 seconds to ~25 seconds. For comparison bootstrapping the "edgeql_select" tests takes about 23 seconds. So I think this cut down the bootstrap time to about as low as it gets. P.S. That's bootstrap without any caching, etc. |
Instead of generating a large dataset to force usage of indexes, disable sequential scan and use a smaller data set. Fixes #6316
Instead of generating a large dataset to force usage of indexes, disable sequential scan and use a smaller data set.
Fixes #6316