[
https://issues.apache.org/jira/browse/HUDI-3362?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
sivabalan narayanan updated HUDI-3362:
--------------------------------------
Summary: Hudi 0.8.0 cannot rollback CoW table (was: Hudi 0.8.0 cannot
roleback CoW table)
> Hudi 0.8.0 cannot rollback CoW table
> ------------------------------------
>
> Key: HUDI-3362
> URL: https://issues.apache.org/jira/browse/HUDI-3362
> Project: Apache Hudi
> Issue Type: Bug
> Affects Versions: 0.8.0
> Reporter: Istvan Darvas
> Assignee: sivabalan narayanan
> Priority: Critical
> Fix For: 0.11.0
>
> Attachments: hoodie-list.txt, hoodie.zip,
> rollback-on-a-not-damaged-table-SUCCESS.pdf,
> rollback-on-a-not-damaged-table-SUCCESS.txt,
> rollback-on-a-not-damaged-table-over-replace-commit.pdf,
> rollback-on-a-not-damaged-table-over-replace-commit.txt,
> rollback_20220131215514.txt, rollback_log.txt, rollback_log_v2.txt
>
> Original Estimate: 4h
> Remaining Estimate: 4h
>
> Hi Guys,
>
> Environment: AWS EMR 6.4 / Hudi v0.8.0
> Problem: I have a CoW table wich is ingested by DeltaStremer (batch style:
> every 5 minutes from Kafka), and after a certain time, DeltaStremer stops
> working with a message like this:
>
> {{diagnostics: User class threw exception:
> org.apache.hudi.exception.HoodieRollbackException: Found commits after time
> :20220131215051, please rollback greater commits first}}
>
> hudi:iot_raw.ingress_pkg_decoded_rep->desc
> ╔════════════════════════════════╤═════════════════════════════════════════════════════════════╗
> ║ Property │ Value
> ║
> ╠════════════════════════════════╪═════════════════════════════════════════════════════════════╣
> ║ basePath │
> s3://scgps-datalake/iot_raw/ingress_pkg_decoded_rep ║
> ╟────────────────────────────────┼─────────────────────────────────────────────────────────────╢
> ║ metaPath │
> s3://scgps-datalake/iot_raw/ingress_pkg_decoded_rep/.hoodie ║
> ╟────────────────────────────────┼─────────────────────────────────────────────────────────────╢
> ║ fileSystem │ s3
> ║
> ╟────────────────────────────────┼─────────────────────────────────────────────────────────────╢
> ║ hoodie.table.name │ iot_raw.ingress_pkg_decoded_rep
> ║
> ╟────────────────────────────────┼─────────────────────────────────────────────────────────────╢
> ║ hoodie.table.type │ COPY_ON_WRITE
> ║
> ╟────────────────────────────────┼─────────────────────────────────────────────────────────────╢
> ║ hoodie.archivelog.folder │ archived
> ║
> ╟────────────────────────────────┼─────────────────────────────────────────────────────────────╢
> ║ hoodie.table.base.file.format │ PARQUET
> ║
> ╟────────────────────────────────┼─────────────────────────────────────────────────────────────╢
> ║ hoodie.timeline.layout.version │ 1
> ║
> ╟────────────────────────────────┼─────────────────────────────────────────────────────────────╢
> ║ hoodie.table.version │ 1
> ║
> ╚════════════════════════════════╧═════════════════════════════════════════════════════════════╝
>
> It is usually a replace commit, I would say I am pretty sure in this.
> I have commits in the timeline:
>
> 20220131214354<-before
> 20220131215051<-error message
> 20220131215514<-after
>
> So as it was told to me, I try to rollback with the following steps in
> hudi-cli:
> 1.) connect --path s3://scgps-datalake/iot_raw/ingress_pkg_decoded_rep /
> SUCCESS
> 2.) savepoint create --commit 20220131214354 --sparkMaster local[2] / SUCCESS
> 3.) savepoint rollback --savepoint 20220131214354 --sparkMaster local[2] /
> FAILED
> 4.) savepoint create --commit 20220131215514 --sparkMaster local[2] / SUCCESS
> 5.) savepoint rollback --savepoint 20220131215514 --sparkMaster local[2] /
> FAILED
>
> Long story short, if I run a situation like this I am not able to solve it
> with the known methods ;) - My use-case is working progress, but I cannot go
> prod with an issue like this.
>
> My question, what would be the right steps / commands to solve an issue like
> this, and be able to restart deltastremer again.
>
> This table, does not have dimension data, so I am happy to share the whole
> table if someone curiuous (if that is needed or would be helpful, lets talk
> in a private mail / slack about the sharing). ~15GB ;) it was stoped after a
> few run, actually after the 1st clustering.
>
> I use this clustering config in the DeltaStremer:
> hoodie.clustering.inline=true
> hoodie.clustering.inline.enabled=true
> hoodie.clustering.inline.max.commits=36
> hoodie.clustering.plan.strategy.sort.columns=correlation_id
> hoodie.clustering.plan.strategy.daybased.lookback.partitions=7
> hoodie.clustering.plan.strategy.target.file.max.bytes=268435456
> hoodie.clustering.plan.strategy.small.file.limit=134217728
> hoodie.clustering.plan.strategy.max.bytes.per.group=671088640
>
> I hope there is someone who can help me to tackle with this, becase if I able
> to solve this manually, I would be confident to go prod.
> So thanks in advance,
> Darvi
> Slack Hudi: istvan darvas / U02NTACPHPU
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
