You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
107297: storage,kvserver: Foundational changes for disaggregated ingestions r=sumeerbhola a=itsbilal
This change contains two commits (split off from the original mega-PR, cockroachdb#105839). The first is a pkg/storage change to add new interface methods to call pebble's db.ScanInternal as well as implement related helper methods in sstable writers/batch readers/writers to be able to do disaggregated snapshot ingestion. The second is a kvserver/rditer change to allow finer-grained control on what replicated spans we iterate on, as well as to be able to specifically opt into skip-shared iteration over the user key span through the use of `ScanInternal`.
---
**storage: Update Engine/Reader/Writer interfaces for ScanInternal**
This change updates pkg/storage interfaces and implementations to allow
the use of ScanInternal in skip-shared iteration mode as well as
writing/reading of internal point keys, range dels and range keys.
Replication / snapshot code will soon rely on these changes to
be able to replicate internal keys in higher levels plus metadata
of shared sstables in lower levels, as opposed to just observed
user keys.
Part of cockroachdb#103028
Epic: none
Release note: None
**kvserver: Add ability to filter replicated spans in Select/Iterate**
This change adds the ability to select for just the replicated
span in rditer.Select and rditer.IterateReplicaKeySpans. Also
adds a new rditer.IterateReplicaKeySpansShared that does a
ScanInternal on just the user key span, to be able to collect
metadata of shared sstables as well as any internal keys above
them.
We only use skip-shared iteration for the replicated user key span
of a range, and in practice, only if it's a non-system range.
Part of cockroachdb#103028.
Epic: none
Release note: None
108336: sql: retry more distributed errors as local r=yuzefovich a=yuzefovich
This PR contains a couple of commits that increase the allow-list of errors that are retried locally. In particular, it allows us to hide some issues we have around using DistSQL and shutting down SQL pods.
Fixes: cockroachdb#106537.
Fixes: cockroachdb#108152.
Fixes: cockroachdb#108271.
Release note: None
108406: server,testutils: remove complexity r=yuzefovich,herkolategan a=knz
There is a saying (paraphrasing) that it always takes more work removing unwanted complexity than it takes to add it. This is an example of that.
Prior to this commit, there was an "interesting" propagation of the flag that decides whether or not to define a test tenant for test servers and clusters. In a nutshell, we had:
- an "input" flag in `base.TestServerArgs`, which remained mostly immutable
- a boolean decided once by `ShouldStartDefaultTestTenant()` either in:
- `serverutils.StartServerOnlyE`
- or `testcluster.Start`
- that boolean choice was then propagated to `server.testServer` via _another_ boolean config flag in `server.BaseConfig`
- both the 2nd boolean and the original input flag were then again checked when the time came to do the work (in `maybeStartDefaultTestTenant`).
Additional complexity was then incurred by the need of `TestCluster` to make the determination just once (and not once per server).
This commit cuts through all the layers of complexity by simply propagating the choice of `ShouldStartDefaultTestTenant()` back into the `TestServerArgs` and only ever reading from that subsequently.
Release note: None
Epic: CRDB-18499
108465: cloudccl: allow external connection tests to be run in parallel r=rhu713 a=rhu713
Currently external connection tests read and write to the same path in cloud storage. Add a random uint64 as part of the path so that test runs have unique paths and can be run in parallel.
Fixes: cockroachdb#107407
Release note: None
108481: acceptance: stabilize start-single-node in tcl test r=santamaura a=dhartunian
We've continued to see flakes on this test which contain messages of throttled stores on node startup. The hypothesis is that these are due to leftover data directories from prior startups during the same test.
This change clears the `logs/db` data directory for those invocations and also adds the sql memory flag which the common tcl function also uses.
Resolvescockroachdb#108405
Epic: None
Release note: None
108496: kv: unit test `PrepareTransactionForRetry` and `TransactionRefreshTimestamp` r=miraradeva a=nvanbenschoten
Informs cockroachdb#104233.
This commit adds a pair of new unit tests to verify the behavior of `PrepareTransactionForRetry` and `TransactionRefreshTimestamp`. These functions will be getting more complex for cockroachdb#104233, so it will be helpful to have these tests in place. The tests also serve as good documentation.
Release note: None
Co-authored-by: Bilal Akhtar <[email protected]>
Co-authored-by: Yahor Yuzefovich <[email protected]>
Co-authored-by: Raphael 'kena' Poss <[email protected]>
Co-authored-by: Rui Hu <[email protected]>
Co-authored-by: David Hartunian <[email protected]>
Co-authored-by: Nathan VanBenschoten <[email protected]>
0 commit comments