James, thank you very much for your fast reply.
I set the saltBuckets to 0, I disabled and dropped the SYSTEM.SEQUENCE, I restartet HBase and reconnected the Phoenix sqlline. After that, it creates SYSTEM.STATS, SYSTEM.FUNCTION and again, SYSTEM.SEQUENCE with many many regions until all region servers crash.

-- Johannes

Am 22.02.2016 um 09:10 schrieb James Taylor:
Hi Johannes,

To override this behavior, you can do the following:
- set phoenix.sequence.saltBuckets in your client-side hbase-site.xml to 0. - manually disable and drop SYSTEM.SEQUENCE from an HBase shell. Note - this assumes you're not using sequences - if you are, let us know.
- re-connect a Phoenix client to the cluster

At this point, the SYSTEM.SEQUENCE table will again be in a single region. FWIW, we've change the default value of the phoenix.sequence.saltBuckets to 0 (I believe as of 4.6), as the previous value was too high for smaller clusters.

Thanks,
James

On Sun, Feb 21, 2016 at 11:57 PM, Johannes <[email protected] <mailto:[email protected]>> wrote:

    Hi,

    I am using the Cloudera Manager on a six-node cluster (master:
    aaa, the others: node[1-5]). It runs CDH 5.5.2 and consists of
    HBase, HDFS and ZooKeeper. I downloaded, distributed and enabled
    the Phoenix Parcel 4.5.2-1.clabs_phoenix1.2.0.p0.774

    On the HBase-Web-UI:
    * Region Servers: my five nodes, two have Num. Regions = 1
    (hbase:namespace and hbase:meta), the others have 0

    Now I start Phoenix:
    * cd /etc/hbase/conf
    * /usr/bin/phoenix-sqlline.py localhost:2181:/hbase

    On the HBase-Web-UI:
    * SYSTEM.CATALOG is created
    * SYSTEM.SEQUENCE is created and has 12 regions
    * Browser reload (F5): Now, SYSTEM.SEQUENCE has 200 region
    * When SYSTEM.SEQUENCE has 224 regions, all Region Servers are dead.
    * Regions in transition. Red entry:
    SYSTEM.SEQUENCE,,1455886910318.84b30d86ec2535f7ff514b9eca1a62ff.
    state=FAILED_OPEN, ts=Fri Feb 19 14:01:52 CET 2016 (34s ago),
    server=aaa,60020,1455886812045

    I also tried to use the Phoenix tar instead of the Cloudera
    parcel. Exactly the same behavior.

    Thanks for every kind of help!

    Johannes



Reply via email to