You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@cassandra.apache.org by Dongsheng Song <do...@gmail.com> on 2025/11/18 10:30:46 UTC

storage failure behavior of Cassandra

Hi,

When Cassandra encountered data persistence failure (e.g.,  storage
pauses/IO hanging, or full storage) for a short time, if we configured
sufficient memory for Cassandra  (e.g., Heap Memory, Direct Memory)
and appropriately increase `memtable_heap_space` and
`memtable_offheap_space`, so that Cassandra has enough memory to save
the memtable even if flushing to SSTable fails, can Cassandra still
provide services and perform data query or modification operations?

Thanks,
Dongsheng

Re: storage failure behavior of Cassandra

Posted by Patrick McFadin <pm...@gmail.com>.

> On Nov 18, 2025, at 2:30 AM, Dongsheng Song <do...@gmail.com> wrote:
> 
> Hi,
> 
> When Cassandra encountered data persistence failure (e.g.,  storage
> pauses/IO hanging, or full storage) for a short time, if we configured
> sufficient memory for Cassandra  (e.g., Heap Memory, Direct Memory)
> and appropriately increase `memtable_heap_space` and
> `memtable_offheap_space`, so that Cassandra has enough memory to save
> the memtable even if flushing to SSTable fails, can Cassandra still
> provide services and perform data query or modification operations?
> 
> Thanks,
> Dongsheng


What you are talking about will keep the memtable flush from happening, but you are missing a key part of durability. The node has to write to the commit log before acknowledging the write.  If the write can’t happen, the node returns a failure. 

This is why replicas are an important part of the uptime feature of Cassandra. If one node is having an intermittent issue, the cluster should work around the troubled node and recover automatically. 

Patrick