Hi Team,
While running the job in application mode...
# troubleshootingc
chunilal kukreja
02/24/2023, 8:21 AMHi Team,
While running the job in application mode in k8s cluster, I m facing an issue that last saved checkpoint is getting restored every time whenever I delete & create a new flink cluster.
Ideally when I am taking a savepoint & starting. the cluster w/o providing any savepoint config details, then how come it restores the last saved checkpoint?
g
Gyula F贸ra
02/24/2023, 9:26 AMAre you using the Flink Kubernetes Operator?
Gyula F贸ra
02/24/2023, 9:27 AMif not, you are likely hitting a problem where you delete your cluster but the HA metadata is kept. In that case I suggest checking the operator which is designed to solve these requirements in a very nice and robust way
馃憦 1
c
chunilal kukreja
02/27/2023, 11:07 AMthanks for pointing out the issue, using k8s operator is nt an option fr us, can you let me knw what needs to be done to solve this HA metadata problem?
g
Gyula F贸ra
02/27/2023, 11:08 AMyou can look at how the operator implements stateful job upgrades
a
Amenreet Singh Sodhi
02/27/2023, 11:10 AMSure, If you have some references or any document which we could refer for this, It would be very helpful for us.
g
Gyula F贸ra
02/27/2023, 11:11 AMGyula F贸ra
02/27/2023, 11:11 AMFinding a way to use the operator will be much easier
Gyula F贸ra
02/27/2023, 11:11 AMyou will save countless hours of sweat
Gyula F贸ra
02/27/2023, 11:11 AMand end up with a much more robust solution
a
Amenreet Singh Sodhi
02/27/2023, 11:13 AMAlso, one more thing, whenever we restart the job, it tries to save checkpointing data to a file and throws the following error:
Copy code
Target file file:/opt/flink/pm/checkpoint/000000006e6b13320000000000000000/chk-1/_metadata already exists.Amenreet Singh Sodhi
02/27/2023, 11:14 AMSure, will see if we could use operator. Thanks