[
https://issues.apache.org/jira/browse/IGNITE-19466?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sergey Uttsel updated IGNITE-19466:
-----------------------------------
Description:
h3. *Motivation*
After introducing data nodes from DistributionZoneManager on table creation it
became possible that a table is created with an empty data nodes value and
therefore with empty assignments. It causes an exceptions like
{code:java}
Caused by: java.lang.AssertionError
at
org.apache.ignite.internal.raft.RaftGroupServiceImpl.randomNode(RaftGroupServiceImpl.java:680)
at
org.apache.ignite.internal.raft.RaftGroupServiceImpl.randomNode(RaftGroupServiceImpl.java:667)
at
org.apache.ignite.internal.raft.RaftGroupServiceImpl.refreshLeader(RaftGroupServiceImpl.java:222)
at
org.apache.ignite.internal.raft.RaftGroupServiceImpl.start(RaftGroupServiceImpl.java:178)
{code}
because there are an empty peers on RaftGroupServiceImpl#start. The proper
solution when a table is created with empty peers but remains unusable until
the data nodes for the zone is updated and RaftGroupServiceImpl is started with
peers.
h3. *Definition of Done*
If the date nodes for the zone are empty, then create a table without a raft
group.
Start raft groups and other components when the data nodes value is updated.
h3. *Implementation Notes*
We have two issues.
# The first is a table creation when the data nodes values of the zone is
empty. In this case createTablePartitionsLocally method gets an empty
newPartAssignment. So it is not possible to do startRaftGroupNode without
peers. Need to skip it if the data nodes is empty.
# The second is starting raft groups when data nodes value is updated. In this
case handleChangePendingAssignmentEvent gets an empty stableAssignments and not
empty pendingAssignments. Currently handleChangePendingAssignmentEvent invokes
startRaftGroupNode to start raft group node with stable assignment and then
invokes changePeersAsync with pendingAssignments. We need to skip it if
stableAssignments is empty and start new raft group with pendingAssignments.
was:
h3. *Motivation*
After introducing data nodes from DistributionZoneManager on table creation it
became possible that a table is created with an empty data nodes value and
therefore with empty assignments. It causes an exceptions like
{code:java}
Caused by: java.lang.AssertionError
at
org.apache.ignite.internal.raft.RaftGroupServiceImpl.randomNode(RaftGroupServiceImpl.java:680)
at
org.apache.ignite.internal.raft.RaftGroupServiceImpl.randomNode(RaftGroupServiceImpl.java:667)
at
org.apache.ignite.internal.raft.RaftGroupServiceImpl.refreshLeader(RaftGroupServiceImpl.java:222)
at
org.apache.ignite.internal.raft.RaftGroupServiceImpl.start(RaftGroupServiceImpl.java:178)
{code}
because there are an empty peers on RaftGroupServiceImpl#start. The proper
solution when a table is created with empty peers but remains unusable until
the data nodes for the zone is updated and RaftGroupServiceImpl is started with
peers.
h3. *Definition of Done*
If the date nodes for the zone are empty, then create a table without a raft
group.
> Empty data nodes on table creation
> ----------------------------------
>
> Key: IGNITE-19466
> URL: https://issues.apache.org/jira/browse/IGNITE-19466
> Project: Ignite
> Issue Type: Bug
> Reporter: Sergey Uttsel
> Priority: Major
> Labels: ignite-3
>
> h3. *Motivation*
> After introducing data nodes from DistributionZoneManager on table creation
> it became possible that a table is created with an empty data nodes value and
> therefore with empty assignments. It causes an exceptions like
> {code:java}
> Caused by: java.lang.AssertionError
> at
> org.apache.ignite.internal.raft.RaftGroupServiceImpl.randomNode(RaftGroupServiceImpl.java:680)
> at
> org.apache.ignite.internal.raft.RaftGroupServiceImpl.randomNode(RaftGroupServiceImpl.java:667)
> at
> org.apache.ignite.internal.raft.RaftGroupServiceImpl.refreshLeader(RaftGroupServiceImpl.java:222)
> at
> org.apache.ignite.internal.raft.RaftGroupServiceImpl.start(RaftGroupServiceImpl.java:178)
> {code}
> because there are an empty peers on RaftGroupServiceImpl#start. The proper
> solution when a table is created with empty peers but remains unusable until
> the data nodes for the zone is updated and RaftGroupServiceImpl is started
> with peers.
> h3. *Definition of Done*
> If the date nodes for the zone are empty, then create a table without a raft
> group.
> Start raft groups and other components when the data nodes value is updated.
> h3. *Implementation Notes*
> We have two issues.
> # The first is a table creation when the data nodes values of the zone is
> empty. In this case createTablePartitionsLocally method gets an empty
> newPartAssignment. So it is not possible to do startRaftGroupNode without
> peers. Need to skip it if the data nodes is empty.
> # The second is starting raft groups when data nodes value is updated. In
> this case handleChangePendingAssignmentEvent gets an empty stableAssignments
> and not empty pendingAssignments. Currently
> handleChangePendingAssignmentEvent invokes startRaftGroupNode to start raft
> group node with stable assignment and then invokes changePeersAsync with
> pendingAssignments. We need to skip it if stableAssignments is empty and
> start new raft group with pendingAssignments.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)