Expose top level stats in scans #227
Conversation
nicklan
force-pushed
the
expose-stats
branch
2 times, most recently
from
20355e6
to
f4445b6
Compare
kernel/src/scan/state.rs
Outdated
| pub struct Stats { | ||
| pub num_records: u64, | ||
| #[serde(default = "default_true")] | ||
| pub tight_bounds: bool, |
There was a problem hiding this comment.
How does it help an engine to know about tight bounds? I guess it tells whether the logical rowcount could be less than the reported physical rowcount, but I wouldn't ever expect a deletion vector to suppress more than a small fraction of rows?
Also: Where do we draw the line on what stats to expose? I guess we stop here because the min/max/nullcount fields are (a) difficult to expose because schema-specific; and (b) arguably it's kernel's job to deal with them, not engine's?
There was a problem hiding this comment.
You're right. I had just started looking at the stats stuff when I wrote this PR, and thought that tight_bounds had implications for the value of this stat, but it doesn't. Rather, the presence of a deletion vector matters.
So I've removed that field, and added a has_vector() -> bool method on DvInfo, and noted what the stat means in a doc comment.
There was a problem hiding this comment.
can we call it has_dv or has_deletion_vector to be self documenting?
Co-authored-by: Ryan Johnson <scovich@users.noreply.github.com>
| const DvInfo* dv_info, | ||
| const CStringMap* partition_values) { | ||
| printf("file: %.*s\n", (int)path.len, path.ptr); | ||
| printf("file: %.*s (size: %" PRId64 ", num_records:", (int)path.len, path.ptr, size); | ||
| if (stats) { |
There was a problem hiding this comment.
aside: should this code also print dv info?
There was a problem hiding this comment.
I'm considering cffi-test.c "deprecated". As soon as #203 merges, I'll remove cffi-test.c and switch our CI over to run the more complete example there, which does look at the dv info
EngineDatawe produce during scansOnly supports
numRecordsfor the moment, but I've kept this as a struct so we can add future stats without breaking apis