[
https://issues.apache.org/jira/browse/IGNITE-20637?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Kirill Tkalenko updated IGNITE-20637:
-------------------------------------
Description:
We need to take care of recovery for the distributed index construction
mechanism:
* For indexes that are *not* available for read-write:
# Make sure that the corresponding keys are present in the metastore:
*startBuildIndex.<indexId>* and
*partitionBuildIndex.<indexId>.<partitionId_0>*...*partitionBuildIndex.<indexId>.<partitionId_N>*.
*Notest:*
This is easy to implement, since there is an auxiliary key
*startBuildIndex.<indexId>*, which, in its absence, will allow you to
accurately understand that there are no keys for this index in metasore.
# For the build distributed index of a specific partition, delete key
*partitionBuildIndex.<indexId>.<partitionId>* in the metastore.
*Notest:*
This point is probably the most difficult and requires thought before
implementation, since it will most likely require raising a replication group
and rolling up a replication log.
# If there are no keys
*partitionBuildIndex.<indexId>.<partitionId_0>*...*partitionBuildIndex.<indexId>.<partitionId_N>*
in the metastore, execute
*org.apache.ignite.internal.catalog.commands.MakeIndexAvailableCommand*.
* For indexes that are available for read-write:
# Delete key *startBuildIndex.<indexId>* in the metastore if it remains.
was:
We need to take care of recovery for the distributed index construction
mechanism:
* For indexes that are not available for read-write:
# Make sure that the corresponding keys are present in the metastore:
*startBuildIndex.<indexId>* and
*partitionBuildIndex.<indexId>.<partitionId_0>*...*partitionBuildIndex.<indexId>.<partitionId_N>*.
*Notest:*
This is easy to implement, since there is an auxiliary key
*startBuildIndex.<indexId>*, which, in its absence, will allow you to
accurately understand that there are no keys for this index in metasore.
# For the build distributed index of a specific partition, delete key
*partitionBuildIndex.<indexId>.<partitionId>*.
*Notest:*
This point is probably the most difficult and requires thought before
implementation, since it will most likely require raising a replication group
and rolling up a replication log.
> Implement recovery of distributed index building
> ------------------------------------------------
>
> Key: IGNITE-20637
> URL: https://issues.apache.org/jira/browse/IGNITE-20637
> Project: Ignite
> Issue Type: Improvement
> Reporter: Kirill Tkalenko
> Priority: Major
> Labels: ignite-3
> Fix For: 3.0.0-beta2
>
>
> We need to take care of recovery for the distributed index construction
> mechanism:
> * For indexes that are *not* available for read-write:
> # Make sure that the corresponding keys are present in the metastore:
> *startBuildIndex.<indexId>* and
> *partitionBuildIndex.<indexId>.<partitionId_0>*...*partitionBuildIndex.<indexId>.<partitionId_N>*.
> *Notest:*
> This is easy to implement, since there is an auxiliary key
> *startBuildIndex.<indexId>*, which, in its absence, will allow you to
> accurately understand that there are no keys for this index in metasore.
> # For the build distributed index of a specific partition, delete key
> *partitionBuildIndex.<indexId>.<partitionId>* in the metastore.
> *Notest:*
> This point is probably the most difficult and requires thought before
> implementation, since it will most likely require raising a replication group
> and rolling up a replication log.
> # If there are no keys
> *partitionBuildIndex.<indexId>.<partitionId_0>*...*partitionBuildIndex.<indexId>.<partitionId_N>*
> in the metastore, execute
> *org.apache.ignite.internal.catalog.commands.MakeIndexAvailableCommand*.
> * For indexes that are available for read-write:
> # Delete key *startBuildIndex.<indexId>* in the metastore if it remains.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
