* refactor: simplify `QueryChunk` data access
We have only two types for chunks (now that the RUB is gone):
1. In-memory RecordBatches
2. Parquet files
Loads of logic is duplicated in the different `read_filter`
implementations. Also `read_filter` hides a solid amount of logic from
DataFusion, which will prevent certain (future) optimizations. To enable #5897
and to simplify the interface, let the chunks return the data (batches
or metadata for parquet files) directly and let `iox_query` perform the
actual heavy-lifting.
* docs: improve
Co-authored-by: Andrew Lamb <alamb@influxdata.com>
* docs: improve
Co-authored-by: Andrew Lamb <alamb@influxdata.com>
Co-authored-by: Andrew Lamb <alamb@influxdata.com>
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
With #5963 merged, all chunks now provide a summary (even though it may
not contain data for all columns). So let's make it mandatory, which
also removes a few 🙈-style `.except(...)` calls.
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
Use the table summary instead. This allows us to have a single mechanism
that both IOx and DataFusion understand. This basically lifts the "basic
table summary" mechanism that the querier uses to `iox_query` and let
the compactor and ingester use the same mechanism.
While not strictly necessary, simplifying the `QueryChunk[Meta]`
interface helps with #5897.
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
This commit makes use of the partition buffer state machine introduced
in https://github.com/influxdata/influxdb_iox/pull/5943.
This commit significantly changes the buffering, and querying, of data
from a partition, swapping out the existing "DataBuffer" for the new
state machine implementation (itself simplified due to temporary lack of
incremental snapshot generation, see #5944).
This commit simplifies the query path, removing multiple types that
wrapped one-another to pass around various state necessary to perform a
query, with various query functions needing different types or
combinations of types. The query path now operates using a single type
(named "QueryAdaptor") that provides a queryable interface over the set
of RecordBatch returned from a partition.
There is significantly increased testing of the PartitionData itself,
covering data in various states and the ordering of returned RecordBatch
(to ensure correct materialisation of updates). There are also
invariants upheld by the type system / compiler to minimise the
complexities of working with empty batches & states, and many asserts
that ensure (mostly existing!) invariants are upheld.
We basically assume everywhere that a column falls into one of the three
known categories (time, tag, field), so lets encode this in our type
system instead of defining "unknown" as "undefined behavior, may or may
not crash".
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
* feat: have a logical plan that is aware of no-deduplication
* feat: build physical scan plan that does not do deduplication
* chore: cleaup
* test: logical plans for scan with and without deduplication
* chore: clean up and a small refactor
* refactor: remove asserts on plan and rename make enable_deduplication default
* refactor: rename disable_deduplication to enable_deduplication
Databases are NOT required to project chunks (in practice this is only
done by the querier for ingester-based chunks). Instead `iox_query`
should (and already does) add the right stream adapters to project
chunks or to create NULL-columns. Removing the special handling from the
test setup makes it easier to understand and also less likely that
`iox_query` starts to rely on this behavior.
Helps with #5897.
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
- remove generic that is basically unused (`group_potential_duplicates`
is always called w/ `Arc<dyn QueryChunk>`)
- remove half-baked `impl` that is unused
Helps w/ #5897.
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
* refactor: `IOxReadFilterNode` can always accumulate statistics
`IOxReadFilterNode` used to not emit statistics if one chunk has
duplicates or delete predicates. This is wrong (or at least overly
conservative), because the node itself (or the chunks themselves) do NOT
perform dedup or delete predicate filtering. Instead this is done is
done by parent nodes (`DeduplicateExec` and `FilterExec`) and its their
job to propagate statistics correctly.
Helps w/ #5897.
* test: explain setup
Co-authored-by: Andrew Lamb <alamb@influxdata.com>
Co-authored-by: Andrew Lamb <alamb@influxdata.com>
Now that we read throw `ParquetExec`, `ROW_GROUP_READ_SIZE` is no longer
used.
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
Use the proper top-level DataFusion context and register the object
store there.
Note that we still hide the `ParquetExec` behind an opaque record batch
stream. Fixing that is next on my list.
Helps with #5897.
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
* refactor: replace `croaring` with `roaring`
With the read buffer gone, roaring bitmaps are only used to calculate
series sets and these calculations are pretty much possible with the
pure-Rust version. Also I don't deem that that performance-critical
(compared to the roaring bitmaps in the read buffer core).
This removes a bunch of dependencies, mostly because `bindgen` is gone.
This also removes our "croaring architecture detection" hack.
* refactor: replace manual roaring sets with arrow
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
* test: Failing tests for unsupported queries
* fix: Catch unsupported SQL operations and error rather than return nothing
* test: Document a few more error messages that come through DataFusion
* refactor: Extract a Step to make query error tests nicer to read and write
* fix: update tests for new error codes
Co-authored-by: Andrew Lamb <andrew@nerdnetworks.org>
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
* feat: initial step to identify where the projection should be provided
* feat: start getting columns of all expressions
* chore: format
* test: test for the table_chunk_stream
* fix: fix a compile error. Thanks @alamb
* test: full tests for table_chunk_stream
* chore: cleanup
* fix: do not cut any columns in case all fields are needed
* test: add one more test case of reading all columns
* refactor: move code that identify columbs ot push down to a function. Add the use of field_columns
* chore: cleanup
* refactor: make sream_from_batch support empty batches
* chore: cleanup
* chore: fix clippy after auto merge
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
- treat OOM protection as "resource exhausted"
- use `DataFusionError` in more places instead of opaque `Box<dyn Error>`
- improve conversion from/into `DataFusionError` to preserve more
semantics
Overall, this improves our error handling. DF can now return errors like
"resource exhausted" and gRPC should now automatically generate a
sensible status code for it.
Fixes#5799.
* chore: Upgrade to Rust 1.64
* fix: Use iter find instead of a for loop, thanks clippy
* fix: Remove some needless borrows, thanks clippy
* fix: Use then_some rather than then with a closure, thanks clippy
* fix: Use iter retain rather than filter collect, thanks clippy
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
* refactor: concurrent table scan in "field columns"
Similar to #5647 and #5649.
* docs: improve
Co-authored-by: Andrew Lamb <andrew@nerdnetworks.org>
Co-authored-by: Andrew Lamb <andrew@nerdnetworks.org>
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
* refactor: concurrent table planning in InfluxRPC
Some InfluxRPC can scan multiple tables. Prior to this PR we were always
scanning the tables in sequence, adding up potential latencies (catalog,
ingester, object store). There is no reason we need to do this,
"ordinary" SQL queries would not serialize this way either.
So let's scan tables concurrently. This add concurrency to:
- read filter
- read group
- read window aggregate
There are other query types that could benefit from a similar treatment.
They will be changed in a follow-up.
* docs: improve
Co-authored-by: Andrew Lamb <andrew@nerdnetworks.org>
* test: explain `Send` assertion
* refactor: change `CONCURRENT_TABLE_JOBS` to 10
Co-authored-by: Andrew Lamb <andrew@nerdnetworks.org>
Only store context, settings (if any) and the schema interner within the
de-duplicator. Extract a new `Chunks` type that handles the chunk
classification and can passed around in a somewhat clean fashion.
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
In our data model, a chunk always belongs to a partition[^1], so let's
not make this attribute optional. The optional value only leads to
-- mostly surprising -- conditional behavior, ranging from "do not equalize
the partition sort key" (querier) to "always consider the chunk overlapping"
(iox_query when dealing with ingester chunks).
[^1]: This is even true when the chunk belongs to a parquet file that is not
yet added to the catalog, contrary to what a comment in the ingester
stated. The catalog and data model used by the querier are two totally
different things.
* fix: apply selection in `TestChunk::read_filter`
TBH I have no idea how this worked so well before, but the chunks are
expected to apply the given selection. This is because
`IOxReadFilterNode::execute` will wrap the `QueryChunk::read_filter`
output into a `SchemaAdapterStream` and this one expects that there are
no input columns that are absent in the output schema (i.e. it will only
add null columns, it won't remove any). Funnily the `SchemaAdapterStream`
error will blame DataFusion for the mess.
* test: make `test_storage_rpc_tag_values_grouped_by_measurement_and_tag_key` a bit harder
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
* refactor: remove dead code
* refactor: `Deduplicator::build_scan_plan` consumes `self`
There is no good reason to use the same `Deduplicator` twice. In
contrast I'm quite sure that this would lead to nasty bugs, because
`split_overlapped_chunks` exists early in some cases so the 2nd plan
would have old and new chunks mixed together.
* ci: use same feature set in `build_dev` and `build_release`
* ci: also enable unstable tokio for `build_dev`
* chore: update tokio to 1.21 (to fix console-subscriber 0.1.8
* fix: "must use"
* fix: hoist repeated computation out of chunk creation
We have hundreds of chunks per table, so it is beneficial to only
do common work once.
* chore: remove TableCache as it is no longer used
* fix: prune chunks both before and after metadata fetch
Fetching the metadata for all the chunks in a table is expensive,
especially when we have a narrow time range query that only
needs a few chunks.
* chore: fix clippy
* fix: fix up some last tests
* fix: review comments
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
Marco Neumann
Andrew Lamb
Dom Dwyer
Carol (Nichols || Goulding)
Nga Tran
kodiakhq[bot]
Carol (Nichols || Goulding)
Andrew Lamb
dependabot[bot]
YIXIAO SHI
Sam Arnold