sql: clean up mutable not-null columns hack #74922
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
This change is |
yuzefovich
reviewed
Jan 18, 2022
There was a problem hiding this comment.
LGTM, but I'll defer to @postamar for approval.
Reviewed 4 of 4 files at r1, all commit messages.
Reviewable status:complete! 0 of 0 LGTMs obtained (waiting on @postamar and @RaduBerinde)
-- commits, line 6 at r1:
nit: s/market/marked/.
-- commits, line 12 at r1:
nit: missing closing parenthesis.
pkg/sql/catalog/tabledesc/column.go, line 318 at r1 (raw file):
c.nonDrop = c.public } else { //readableDescs := make([]descpb.ColumnDescriptor, 0, numMutations)
Should this be now removed?
Mutation columns in some cases need to be scanned even if they haven't been backfilled yet, which means that we may retrieve NULL values even if they are marked as not-nullable. We currently have a hack in the table descriptor which changes the nullable flags in the column descriptors when `ReadableColumns()` is used. It is very surprising that we can get different descriptors for a given ColumnID depending if we look for it in `ReadableColumns()` or in `AllColumns()` (e.g. via FindColumnWithID). This commit cleans this up, changing the scanning code to check for `Public()` instead. Release note: None
RaduBerinde
force-pushed
the
remove-readable-hack
branch
from
b2e4fb5 to
21bace2
Compare
RaduBerinde
commented
Jan 18, 2022
There was a problem hiding this comment.
Reviewable status:
pkg/sql/catalog/tabledesc/column.go, line 318 at r1 (raw file):
Previously, yuzefovich (Yahor Yuzefovich) wrote…
Should this be now removed?
Oops, done.
fqazi
approved these changes
Jan 19, 2022
There was a problem hiding this comment.
le descriptor which changes the
nullable flags in the column descr
LGTM. Much cleaner than the original hacky code.
Reviewed 1 of 1 files at r2, all commit messages.
Reviewable status:
fqazi
approved these changes
Jan 19, 2022
There was a problem hiding this comment.
Much cleaner than the original hacky code.
Reviewable status:
|
TFTR! bors r+ |
|
Build succeeded: |
|
Thanks for doing this! |
craig bot
pushed a commit
that referenced
this pull request
Jan 21, 2022
74318: tracing: add /debug/tracez rendering the active spans r=andreimatei a=andreimatei `/debug/tracez` lets users take a snapshot of the active spans registry and render the new snapshot, or one of the previously taken snapshots. The Tracer can hold up to 10 snapshots in memory. It looks like this:  When visualizing a snapshot, the page lets you do a number of things: 1. List all the spans. 2. See the (current) stack trace for each span's goroutine (if the goroutine was still running at the time when the snapshot was captured). Stack traces can be toggled visible/hidden. 3. Sort the spans by name or start time. 4. Filter the span according to text search. The search works across the name and stack trace. 5. Go from a span to the full trace containing that span. For the table Javascript providing sorting and filtering, this patch embeds the library from https://listjs.com/ . Limitations: - for now, only the registry of the local node is snapshotted. In the fuiture I'll collect info from all nodes. - for now, the relationships between different spans are not represented in any way. I'll work on the ability to go from a span to the whole trace that the span is part of. - for now, tags and structured and unstructured log messages that a span might have are not displayed in any way. At the moment, span creation is not enabled in production by default (i.e. the Tracer is put in `TracingModeOnDemand` by default, instead of the required `TracingModeActiveSpansRegistry`). This patch does not change that, so in order to benefit from /debug/tracez in all its glory, one has to run with `COCKROACH_REAL_SPANS=1` for now. Not for long, though. Release note: None 74867: sql: Support CREATE DATABASE WITH OWNER r=Fenil-P a=Fenil-P fixes #67817 Release note (sql change): Allow users to specify the owner when creating a database. Similar to postgresql: CREATE DATABASE name [ [ WITH ] [ OWNER [=] user_name ] 74871: sql: add a tracing tag with the txn ID r=andreimatei a=andreimatei This patch adds the txn's ID as a tag to the tracing span representing a SQL txn. I'm creating a UI to explore the current spans, and this ID will make it easy to navigate between a query/request blocking on a lock held by some other txn, and the activity of that other txn. Release note: None 75114: sql: directly specify columns in TableReader r=RaduBerinde a=RaduBerinde ~Note: the first commit is #74922.~ The internal columns of the TableReader (as well as the row fetcher) are all the columns of the table, with only a subset of values actually produced. This design decision has been carried over way past the point where it makes sense (I admit, it's questionable whether it ever made sense). For one, "all the columns" is ambiguous (does it contain non-public columns? does it include system columns?) leading to various flags and inherent fragility. Second, it relies on the execution engine to figure out (based on the PostProcessSpec) which columns are actually needed, which the optimizer already figures out for us now. This commit changes the TableReader spec and the interface of row.Fetcher to always produce a given specific set of column IDs. The diagram for table readers now specifies the columns by name. The JoinReader, InvertedJoiner, ZigzagJoiner are not changed in this commit (but they should be cleaned up as well). Release note: None 75175: colfetcher: fix the bytes read statistic collection r=yuzefovich a=yuzefovich During 21.2 release we adjusted the `cFetcher` to be `Close`d eagerly when it is returning the zero-length batch. This was done in order to release some references in order for the memory to be GCed sooner; additionally, the `cFetcher` started being used for the index join where the fetcher is restarted from scratch for every batch of spans, so it seemed reasonable to close it automatically. However, that eager closure broke "bytes read" statistic collection since the `row.KVFetcher` was responsible for providing it, and we were zeroing it out. This commit fixes this problem by the `cFetcher` memorizing the number of bytes it has read in `Close`. Some care needs to be taken to not double-count the bytes read in the index join, so a couple of helper methods have been introduced. Additionally this commit applies the same eager-close optimization to the `cFetcher` when the last batch is returned (which makes it so that if we've just exhausted all KVs, we close the fetcher - previously, we would set the zero length on the batch and might never get into `stateFinished`). Fixes: #75128. Release note (bug fix): Previously, CockroachDB could incorrectly report `KV bytes read` statistic in `EXPLAIN ANALYZE` output. The bug is present only in 21.2.x versions. 75215: cmd/github-post: fix Pebble metamorphic reproduction command r=jbowens a=jbowens When posting a github issue for a Pebble metamorphic test failure, include the correct `-ops` flag. Discovered because cockroachdb/pebble#1459 contained a reproduction command that contained too few ops to reproduce the issue. Release note: none 75228: logictestccl: skip flaky TestCCLLogic/fakedist-metadata/partitioning_enum r=mgartner a=mgartner Informs #75227 Release note: None 75237: cli,rpc: don't check the active cluster version in the CLI r=andreimatei a=knz This commit removes a code path that would tickle an assertion failure if we were to later fix the context propagation in the RPC heartbeat method (see PR #71243): there's no "active cluster version" in the CLI and so we can't compare it in a client interceptor. Release note: None 75254: scripts: add `dev generate --mirror` to `bump-pebble.sh` script r=jbowens a=nicktrav CI now expects that dependencies are mirrored to cloud storage and will fail if the TODO for mirroring the repo is left unaddressed in the `DEPS.bzl` file. Add a mirroring step to the `bump-pebble.sh` script. Release note: none Co-authored-by: Andrei Matei <[email protected]> Co-authored-by: Fenil Patel <[email protected]> Co-authored-by: Radu Berinde <[email protected]> Co-authored-by: Yahor Yuzefovich <[email protected]> Co-authored-by: Jackson Owens <[email protected]> Co-authored-by: Marcus Gartner <[email protected]> Co-authored-by: Raphael 'kena' Poss <[email protected]> Co-authored-by: Nick Travers <[email protected]>
10 tasks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Mutation columns in some cases need to be scanned even if they haven't
been backfilled yet, which means that we may retrieve NULL values even
if they are marked as not-nullable.
We currently have a hack in the table descriptor which changes the
nullable flags in the column descriptors when
ReadableColumns()isused. It is very surprising that we can get different descriptors for
a given ColumnID depending if we look for it in
ReadableColumns()orin
AllColumns()(e.g. via FindColumnWithID).This commit cleans this up, changing the scanning code to check for
Public()instead.Release note: None