[
https://issues.redhat.com/browse/ISPN-11129?page=com.atlassian.jira.plugi...
]
Gustavo Fernandes updated ISPN-11129:
-------------------------------------
Description:
'Non-shared indexes' is a setup when each node has a local index backed by the
file system, using either the near-real-time or the FS index managers.
For REPL caches, a full copy of the index is kept in each node. As long as at least on
node is up the queries hit the local index which represents the full index of the data.
But non-shared indexes has some drawbacks for DIST caches: each node only indexes data
that the node is the primary owner. If a node goes down, there is no redundancy. This
issue aims to solve this issue at two levels:
* Indexing: replicated entries should be indexed for redundancy purposes. If a primary
owner goes down, the entry is indexed in the backup owner. This should follow exactly the
data distribution on the cache in terms of number of replicas and rebalancing
* Querying: query broadcast is currently used for non-shared indexes. If replicated
entries start to be indexed, a mechanism will have to be used to avoid retuning repeated
results, since currently the broadcast simply queries all the entries in each node and
join the result.
was:
'Non-shared indexes' is a setup when each node has a local index backed by the
file system, using either the near-real-time or the FS index managers.
For REPL caches, a full copy of the index is kept in each node. As long as at least on
node is up the queries hit the local index which represents the full index of the data.
But non-shared indexes has some drawbacks for DIST caches: each node only indexes data
that the node is the primary owner. If a node goes down, there is no redundancy. This
issue aims to solve this issue at two levels:
* Indexing: replicated entries should be indexed for redundancy purposes. If a primary
owner goes down, the entry is indexed in the backup owner. This should follow exactly the
data distribution on the cache in terms of number of replicas and rebalancing
* Querying: query broadcast is currently used for non-shared indexes. If replicated
entires start to be indexed, a mechanism will have to be used to avoid retuning repeated
results, since currently the broadcast simply queries all the entries in each node and
join the result.
High Availability for non-shared indexes on DIST caches
-------------------------------------------------------
Key: ISPN-11129
URL:
https://issues.redhat.com/browse/ISPN-11129
Project: Infinispan
Issue Type: Bug
Components: Embedded Querying
Reporter: Gustavo Fernandes
Assignee: Gustavo Fernandes
Priority: Major
'Non-shared indexes' is a setup when each node has a local index backed by the
file system, using either the near-real-time or the FS index managers.
For REPL caches, a full copy of the index is kept in each node. As long as at least on
node is up the queries hit the local index which represents the full index of the data.
But non-shared indexes has some drawbacks for DIST caches: each node only indexes data
that the node is the primary owner. If a node goes down, there is no redundancy. This
issue aims to solve this issue at two levels:
* Indexing: replicated entries should be indexed for redundancy purposes. If a primary
owner goes down, the entry is indexed in the backup owner. This should follow exactly the
data distribution on the cache in terms of number of replicas and rebalancing
* Querying: query broadcast is currently used for non-shared indexes. If replicated
entries start to be indexed, a mechanism will have to be used to avoid retuning repeated
results, since currently the broadcast simply queries all the entries in each node and
join the result.
--
This message was sent by Atlassian Jira
(v7.13.8#713008)