workload/schemachange: classify and handle errors in all operations #56119
Assignees
Labels
A-testing
Testing tools and infrastructure
C-enhancement
Solution expected to add code/behavior + preserve backward-compat (pg compat issues are exception)
Comments
jayshrivastava
added
C-enhancement
This was referenced Oct 29, 2020
This was referenced Nov 6, 2020
craig bot
pushed a commit
that referenced
this issue
Nov 15, 2020
56184: workload/schemachange: screen for errors in operations r=ajwerner a=jayshrivastava workload/schemachange: screen for errors in operations These changes add error screening for the following (#56119): - createEnum op - createSchema op - dropColumn op - dropColumnDefault op - dropColumnNotNull op - dropSchema op - renameColumn op - renameTable op - renameView op Also, errors caused by schema changes after writes in the same txn are handled (#56230) It also resolves a bug related to ambiguity in column names that was caused by views/tables referencing the same columns (#56235). Co-authored-by: Jayant Shrivastava <[email protected]>
craig bot
pushed a commit
that referenced
this issue
Dec 3, 2020
56379: workload/schemachange: improve error screening r=ajwerner a=jayshrivastava workload/schemachange: improve error screening Several ops will be updated to screen for errors (#56119): - dropTable - dropView - createSequence - renameSequence - dropSequence - insertRow Sequences were also updated to have more interesting cases: non-existing sequences can randomly be returned and sequences can be owned by columns. Furthermore, this change fixes a bug where non-existing tables were getting returned instead of existing ones. Finally, error screening logic is updated to ignore transaction retry errors. This fixes one of the issues in #56230. 57034: cli: add reset-quorum command r=knz a=TheSamHuang This adds `reset-quorum` as a command in the debug CLI. This command utilizes the functionality added in #56333 to recover unavailable ranges where all replicas are lost. 57272: roachtest: Add missing gossiped StoreDescriptor to expected relocate errors r=nvanbenschoten a=nvanbenschoten Closes #57191. This seems possible shortly after the cluster starts up, if gossip hasn't propagated by the time the `ALTER TABLE _ EXPERIMENTAL_RELOCATE _` statement is run. 57274: roachtest: remove initial disk space check from drop test r=nvanbenschoten a=nvanbenschoten Fixes #56040. Will need to be backported. This assertion seemed like a good idea, but it was repeatedly (dec148a) a source of flakiness when fixtures changed. Now that we're using IMPORT for TPC-C, the check is even harder to get right without making it so small as to be useless. It doesn't seem to be worth the instability, so remove it. 57454: util/log: de-flake TestSecondaryGC and some SQL tests r=irfansharif a=knz Fixes #57442 Fixes #57444 See individual commits for details. Co-authored-by: Jayant Shrivastava <[email protected]> Co-authored-by: Sam Huang <[email protected]> Co-authored-by: Nathan VanBenschoten <[email protected]> Co-authored-by: Raphael 'kena' Poss <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Right now errors occur very frequently inside the schemachange workload and there is no mechanism that checks to see if the error makes sense or not. For example, if we try to add a column to a table that does not exist, we should assert that we receive the right error code for an undefined table.
All ops should be updated up to the standard above:
Once all the ops are up to date:
The text was updated successfully, but these errors were encountered: