On Fri, Jun 28, 2019 at 9:47 AM James Kebinger
wrote:
> We were thinking that meta replication would make a hot meta table on a
> busy cluster less hot rather than more hot.
You are running read replicas or replicating the meta to another cluster?
S
> Is the meta replication feature
> for
Hey Stack, thanks for your response
We were thinking that meta replication would make a hot meta table on a
busy cluster less hot rather than more hot. Is the meta replication feature
for uptime rather than performance if it, as you said, will "up the
temperature even more"?
I was surprised that
If early 2.0 or 2.1 versions, perhaps it is HBASE-21292 ?
S
On Thu, Jun 27, 2019 at 10:13 PM Stack wrote:
> On Wed, Jun 26, 2019 at 6:47 AM James Kebinger
> wrote:
>
>> Hello,
>> We're experimenting with meta table replication
>
>
> Replicating the meta table itself? What you thinking?
>
>
>
On Wed, Jun 26, 2019 at 6:47 AM James Kebinger
wrote:
> Hello,
> We're experimenting with meta table replication
Replicating the meta table itself? What you thinking?
> but have found that the
> region servers hosting the replicas will get extremely high priority RPC
> handler usage,
It appears the rpc handlers are getting stuck reading the hfiles backing
the meta table ( based on the block read errors, hedged reads, and
hfilescanner lines in the trace ). I'd check that the backing hdfs cluster
is healthy and consider allocating more I/O capacity for the meta table.
Meta is
Hello,
We're experimenting with meta table replication but have found that the
region servers hosting the replicas will get extremely high priority RPC
handler usage, sometimes all the way to 100% at which point clients start
to experience errors - the priority RPC handler usage is much higher
