[
https://issues.apache.org/jira/browse/IGNITE-20121?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Andrey Mashenkov updated IGNITE-20121:
--------------------------------------
Description:
Index destruction, before starting, must wait until the partition’s SafeTime
becomes >= ‘Activation moment of index removal’ (aka ‘end time of STOPPING
state’ for this index). This is to avoid a race between operations on the index
(including writes and reads) and its destruction.
UPD:
Index can be destroyed with all its data in the next cases
# if an index was dropped in REGISTERED state, it can be destroyed instantly.
# if the last Catalog version, which had the index in STOPPING state, has been
deleted. E.g. after Catalog compaction.
# if an index was dropped in BUILDING state, it can be destroyed right after
removal. I.e. after transition BUILDING->STOPPED->REMOVED.
For the last one, we may need an additional check or index state to decided if
the index can be dropped instantly or should be dropped later.
was:
Index destruction, before starting, must wait until the partition’s SafeTime
becomes >= ‘Activation moment of index removal’ (aka ‘end time of STOPPING
state’ for this index). This is to avoid a race between operations on the index
(including writes and reads) and its destruction.
UPD:
Index can be destroyed with all its data in the next cases
* if an index was dropped in REGISTERED state, it can be destroyed instantly.
* if the last Catalog version, which had the index in STOPPING state, has been
deleted. E.g. after Catalog compaction.
> Postpone index destruction until index is unreachable for RO transactions via
> Catalog
> -------------------------------------------------------------------------------------
>
> Key: IGNITE-20121
> URL: https://issues.apache.org/jira/browse/IGNITE-20121
> Project: Ignite
> Issue Type: Improvement
> Reporter: Roman Puchkovskiy
> Priority: Major
> Labels: ignite-3
> Fix For: 3.0.0-beta2
>
>
> Index destruction, before starting, must wait until the partition’s SafeTime
> becomes >= ‘Activation moment of index removal’ (aka ‘end time of STOPPING
> state’ for this index). This is to avoid a race between operations on the
> index (including writes and reads) and its destruction.
> UPD:
> Index can be destroyed with all its data in the next cases
> # if an index was dropped in REGISTERED state, it can be destroyed instantly.
> # if the last Catalog version, which had the index in STOPPING state, has
> been deleted. E.g. after Catalog compaction.
> # if an index was dropped in BUILDING state, it can be destroyed right after
> removal. I.e. after transition BUILDING->STOPPED->REMOVED.
> For the last one, we may need an additional check or index state to decided
> if the index can be dropped instantly or should be dropped later.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
