Skip to content

v1.18.0-snapshot.20210928.7948.0.b4d00317

v1.18.0-snapshot.20210928.7948.0.b4d00317: detect unsynchronized contract table and retry (#10617)
@S11001001 S11001001 tagged this 28 Sep 20:47 
* enumerating out-of-sync offsets at the DB level

* cleanup in lastOffset

* write the latest-requested-or-read offset when catching up

- Writing only the latest-read, as before, would imply unsynced offsets
  that are actually well-synced.  This puts the DB in a more uniform
  state, i.e. it should actually reflect the single value that the
  fetchAndPersist loop tries to catch everything up to.

* detecting lagging offsets from the unsynced-offsets set

- Treating every unsynced offset as a lag would make us needlessly retry
  perfectly synchronized query results.

* add Foldable1 derived from Foldable for NonEmpty

* nicer version of the unsynced function

* ConnectionIO scalaz monad

* rename Offset.ordering to `Offset ordering` so it can be imported verbatim

* finish aggregating in the lag-detector function, compiles

* port sjd

* XTag, a scalaz 7.3-derived tag to allow stacked tags

* make the complicated aggregation properly testable

* extra semantic corner cases I didn't think of

* tests for laggingOffsets

* a way to rerun queries if the laggingOffsets check reveals inconsistency

* if bookmark is ever different, we always have to rerun anyway

* boolean blindness

* incorporate laggingOffsets into fetchAndPersistBracket

* split fetchAndPersist from getTermination and clean up its arguments

* just compose functors

* add looping to fetchAndPersistBracket

* more mvo tests

* test unsyncedOffsets, too

* Lagginess collector

* supply more likely actual data with mvo tests; don't trust Java equals

* rework minimumViableOffsets to track sync states across template IDs

* extra note

* fix the tests to work against the stricter mvo

* move surrogatesToDomains call

* more tests for lagginess accumulator

* add changelog

CHANGELOG_BEGIN
- [JSON API] Under rare conditions, a multi-template query backed by database
  could have an ACS portion that doesn't match its transaction stream, if
  updated concurrently.  This conditions is now checked and accounted for.
  See `issue #10617 <https://github.com/digital-asset/daml/pull/10617>`__.
CHANGELOG_END

* port toSeq to Scala 2.12

* handle a corner case with offsets being too close to expected values

* didn't need XTag
Assets 2
Loading