ASV performance improvements #2834
Conversation
poodlewars
added
patch
poodlewars
force-pushed
the
aseaton/asv/basic-functions-slowness
branch
from
f9b32c3 to
e19ce1a
Compare
poodlewars
changed the title
… rows is pointless.
…tions information about the setup_cache time is clearer
Remove unused start_time = time.time()
To start with this took 63s on my machine. After the change for string generation, this took 8s. After the change for date generation, this took 3s.
…y run for each "rows" parameter, even though they do not use it.
… ModificationFunctions otherwise they run for each "rows" parameter, even though they do not use it.
poodlewars
force-pushed
the
aseaton/asv/basic-functions-slowness
branch
from
e19ce1a to
8c66918
Compare
| param_names = PARAM_NAMES | ||
|
|
||
| CONNECTION_STRING = "lmdb://basic_functions" | ||
| DATE_RANGE = DATE_RANGE |
| self.fresh_lib._nvs.compact_incomplete(f"sym", False, False) | ||
|
|
||
|
|
||
| class ShortWideWrite: |
There was a problem hiding this comment.
Is there a reason to keep the ShortWideWrite and UltraShortWideWrite separated by have parametrized ShortWideRead
| warmup_time = 0 | ||
| timeout = 6000 | ||
| CONNECTION_STRING = "lmdb://batch_basic_functions?map_size=20GB" | ||
| sample_time = 0.1 |
There was a problem hiding this comment.
What was the process of coming up with this?
| os.remove(self.path) | ||
|
|
||
| def create_dict(self, size): | ||
| ten_char_strings = [random_string(10) for _ in range(1000)] |
There was a problem hiding this comment.
I doubt it will matter much but maybe it'll be a good idea to add some seeds for all the random stuff below
| self.lib.append("sym", self.df_append_single) | ||
|
|
||
| def time_append_large(self, lad: LargeAppendDataModify, rows): | ||
| large: pd.DataFrame = lad.df_append_large[rows].pop(0) |
There was a problem hiding this comment.
lad seems to be used only here. Does it make sense to make a separate benchmark class for it?
| rounds = 1 | ||
| number = 1 # We do a single run between setup and teardown because we e.g. can't delete a symbol twice | ||
| repeat = 2 | ||
| warmup_time = 0 |
There was a problem hiding this comment.
I wonder why it's included in some tests and some doesn't
|
|
||
| class IterateVersionChain: | ||
| timeout = 6000 | ||
| timeout = 1000 |
There was a problem hiding this comment.
I see this has been tweaked a few times. Given that its backend is LMDB so the storage less likely to fail, shall this be relaxed to prioritize outputting result at the end?
No strong opinion on this though
Note that this includes a non-test change I came across while working on this, just to fix the logging when we write empty dataframes, which was malformed. This is e7ecd73.
The other changes are to speed up or otherwise improve our ASV execution.
This has reduced the benchmarking time to roughly 1h40 down from about 2h20m.
Please review commit-by-commit.