m****r 发帖数: 6639 | 1 just wasted a whole day on this shite.
when i created a keyspace, i forgot to set the replication factor. but the
keyspace was created, and shows up in the GUI tool and everything. I can
even create column_families inside the keyspace.
but when i try to insert stuff into the cf's, it stops working without any
meaningful error message. wtf? | c******n 发帖数: 4965 | 2 cassandra's schema replication is flawed by design
you can have schema out of sync,
it could be that when you try to insert into the cf, it went to a
certain node, and that node does not have the latest schema def , and
that cf is not present
I once repeatedly reproduced the schema out of sync issue, but no time
to dig into that further now
but the
can
any
【在 m****r 的大作中提到】 : just wasted a whole day on this shite. : when i created a keyspace, i forgot to set the replication factor. but the : keyspace was created, and shows up in the GUI tool and everything. I can : even create column_families inside the keyspace. : but when i try to insert stuff into the cf's, it stops working without any : meaningful error message. wtf?
| g*****g 发帖数: 34805 | 3 Why do you need to create CF on the fly? For testing purpose, yes.
But I don't see it's needed in production.
【在 c******n 的大作中提到】 : cassandra's schema replication is flawed by design : you can have schema out of sync, : it could be that when you try to insert into the cf, it went to a : certain node, and that node does not have the latest schema def , and : that cf is not present : I once repeatedly reproduced the schema out of sync issue, but no time : to dig into that further now : : but the : can
| c******n 发帖数: 4965 | 4 it's not a matter of "on the fly", if it gets a schema discrepancy in the
matter of seconds, yes, its failure rate in tests are high; in prod that
failure rate could be very very low, but still possible. but if it
occurs, it could be quite serious
【在 g*****g 的大作中提到】 : Why do you need to create CF on the fly? For testing purpose, yes. : But I don't see it's needed in production.
| g*****g 发帖数: 34805 | 5 In most cases, you don't need to create a schema on the fly.
So all the CFs are in every node before you start your app.
And then you can add new node to the cluster and let it replicate.
【在 c******n 的大作中提到】 : it's not a matter of "on the fly", if it gets a schema discrepancy in the : matter of seconds, yes, its failure rate in tests are high; in prod that : failure rate could be very very low, but still possible. but if it : occurs, it could be quite serious
|
|