Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

storage: leave intents behind after blind-writes experiencing write-too-old #44654

Merged
merged 3 commits into from
Feb 13, 2020
Merged

storage: leave intents behind after blind-writes experiencing write-too-old #44654

merged 3 commits into from
Feb 13, 2020

Commits on Feb 13, 2020

  1. storage: fix test stale comment 

    Release note: None
    @andreimatei
    andreimatei committed Feb 13, 2020
    Configuration menu
    Copy the full SHA
    9b84126 View commit details
    Browse the repository at this point in the history

  2. storage: remove obsolete code 

    Remove InclusiveTimeBounds(), a vestige that used to be more complex for
backwards compatibility with 19.1 reasons but became obsolete when we
cleaned up the compatibility cruft.

Release note: None
    @andreimatei
    andreimatei committed Feb 13, 2020
    Configuration menu
    Copy the full SHA
    56cced1 View commit details
    Browse the repository at this point in the history

  3. storage: leave intents behind after blind-writes experiencing write-t… 

    …oo-old

Before this patch, any write running into a write-too-old condition
resulted in a WriteTooOldError being returned by the server. Returning
an error implies that no intents are left behind. This is unfortunate;
we'd like to leave intents (or, in the future, other types of locks)
behind so keep away other transactions. We've observed this resulting in
the starvation of a class of transactions in a user's workload.

This patch makes it so that blind writes (i.e. Puts - used by UPDATE,
not CPuts) don't return WriteTooOldErrors any more. Instead, they return
the a txn proto with the WriteTooOld flag set. This is the behavior they
had before cockroachdb#38668. This patch retains the goal of cockroachdb#38668, however: the
client now eagerly refreshes the transactions when it sees a WriteTooOld
flag, and if the refresh succeeds, it returns a WriteTooOldError to the
higher layers (SQL), allowing for automatic retries where applicable.

Unfortunately, CPuts (used by INSERT) continue to return
WriteTooOldErrors without leaving locks behind. Dealing with them
requires more tenderness because they imply a read, and the timestamp of
a read cannot be bumped as easily as that of a write.

Touches cockroachdb#44653

Release note (sql change): UPDATEs returning a serialization failure error (code
40001) now leave behind a lock, helping the transaction succeed if it
retries. This prevents starvation of transactions whose UPDATEs are
prone to conflicts.
    @andreimatei
    andreimatei committed Feb 13, 2020
    Configuration menu
    Copy the full SHA
    da5dc5b View commit details
    Browse the repository at this point in the history