v0 Datafusion with late materialization #414
Conversation
a10y
force-pushed
the
aduffy/df-pushdown-v0
branch
from
b4fb99c
to
f17cb70
Compare
|
Output of the Note that 
Even though this is synthetic data, it still illustrates that decoding overhead is the driving factor in execution time. There's also some latency between uncompressed Vortex with no pushdown and Arrow with no pushdown, but that time is roughly the ~130µs it takes to do the Vortex -> Arrow conversion (benchmarked that separately, not in the repo). |
|
Right now we don't run the filters on compressed data which would probably be the thing to fix. Anyway, this seems fixable |
|
I agree. I'm going to address the last few comments of your original review and then convert this to "Ready for review" |
a10y
changed the title
a10y
force-pushed
the
aduffy/df-pushdown-v0
branch
from
e5d82a8
to
8cd7f8f
Compare
a10y
force-pushed
the
aduffy/df-pushdown-v0
branch
from
994e73b
to
df06c5f
Compare
vortex-datafusion/src/lib.rs
Outdated
| let filter_exprs: Option<Vec<Expr>> = if filters.is_empty() { | ||
| None | ||
| } else { | ||
| Some(filters.to_vec()) |
There was a problem hiding this comment.
my earlier comment meant that you can get rid of this to_vec
vortex-datafusion/src/lib.rs
Outdated
| @@ -261,7 +483,7 @@ impl ExecutionPlan for VortexMemoryExec { | |||
| self.array.clone() | |||
| }; | |||
|
|
|||
| Self::execute_single_chunk(chunk, &self.projection, context) | |||
| execute_unfiltered(&chunk, &self.scan_projection) | |||
There was a problem hiding this comment.
not sure why this change, you immediately clone the chunk in the function. How about we also change the scan_projection to &[usize]?
vortex-datafusion/src/lib.rs
Outdated
| /// Check if the given expression tree can be pushed down into the scan. | ||
| fn can_be_pushed_down(expr: &Expr, schema_columns: &HashSet<String>) -> DFResult<bool> { | ||
| // If the filter references a column not known to our schema, we reject the filter for pushdown. | ||
| // TODO(aduffy): is this necessary? Under what conditions would this happen? |
There was a problem hiding this comment.
Reading the docs I don't think this check is necessary. TableProvider returns the schema so datafusion will know if the query is valid or not
There was a problem hiding this comment.
i just removed it, if we end up needing to add it back later for some reason we can reference this PR
Unfinished, just opening this as I continue to get things working.
This PR augments the original Vortex connection for Datafusion, with an implementation of filter pushdown that allows us to perform late materialization on as many columns as possible.
Pushdown support will be able to get flagged on/off so we can run benchmarks testing different strategies.
I'm hoping to have an initial version of this with a benchmark harness tonight.