Generally speaking, rejected mutations due to resource contention is
considered a system failure, requiring a re-examination of system resources.
That requires re-architecting your ingest or adding significant resources.
You could do some substantial pre-processing of your ingest and bulk-load
th
@ Eric: yes I have notices 3GB to 5GB swap uses out of 32GB on servers.
And if I will resend the mutations rejected explicitly then this may
create a loop for mutations getting rejected again and again. Then how
can I handle it? How did you? Am i getting it right?
@ Josh: For one of the zookeep
Eric Newton wrote:
Failure to talk to zookeeper is *really* unexpected.
Have you noticed your nodes using any significant swap?
Emphasis on this. Failing to connect to ZooKeeper for 60s (2*30) is a
very long time (although, I think I have seen JVM GC pauses longer before).
A couple of gene
I was simplifying a bit too much. If an error propagates all the way to an
Accumulo client call, then it has stopped retrying for you.
An example:
- create a batchwriter. this creates an update session within the tserver
- mutations are sent against this session id
- mutations are pushed
Thanks for the beautiful explanation Eric, so this means that if I get
Mutations rejected exception due to tablet server failure, the
batchwriter will resend them to some other server and I do not have
worry about them. Great...
But what is the case when we get mutations rejected exception a
By default, accumulo traces major and minor compactions.
Distributed tracing is one way we try to figure out where time is being
spent. You can read the Google Dapper paper to get a better description of
the framework.
The tracing framework pushes the trace information into the trace table by
for
The accumulo batch writer will re-send mutations if a tablet server fails,
or rejects the mutations because the tablet has moved. There's nothing you
have to do to recover from fail-overs and re-balancing.
I'm not a kernel expert, but I believe that a swappiness setting of "1" is
equivalent to "0
And why are there 5000 spans queued for delevery?
*Tracing spans are being dropped because there are already 5000 spans queued
for delivery.
This does not affect performance, security or data integrity, but distributed
tracing information is being lost.*
On 12/23/2015 10:01 AM, mohit.kaushik
I have 3 tablet servers having around 1.4K tablets. If a tablet server
loses its session with zookeeper and killed itself. The system takes
some time to move all hosted tablets to other servers.
In this case if a ingest in process then what should happen with the
mutations going to tablets h
A tablet server is given the rights to manage a tablet.
It is critical that no other server uses the tablet to maintain consistency.
To maintain the right to access a tablet, it must maintain a zookeeper
session. The zookeeper session periodically exchanges keep-alive messages.
If either party fa
10 matches
Mail list logo
