Local indexes are stored in the same table as the data. They are "local"
to the data.
I would not be surprised if you are running into issues because you are
using such an old version of Phoenix.
On 7/24/19 10:35 PM, Alexander Lytchier wrote:
Hi,
We are currently using Cloudera as a package manager for our Hadoop
Cluster with Phoenix 4.7.0 (CLABS_PHOENIX)and HBase 1.2.0-cdh5.7.6.
Phoenix 4.7.0 appears to be the latest version supported
(http://archive.cloudera.com/cloudera-labs/phoenix/parcels/latest/) even
though it’s old.
The table in question has a binary row-key: pk BINARY(30): 1 Byte for
salting, 8 Bytes - timestamp (Long), 20 Bytes - hash result of other
record fields. + 1 extra byte for unknown issue about updating schema in
future (not sure if relevant). We are currently facing performance
issues and are attempting to mitigate it by adding secondary indexes.
When generating a local index synchronously with the following command:
CREATE LOCAL INDEX INDEX_TABLE ON “MyTable” (“cf”.”type”);
I can see that the resulting index table in Phoenix is populated, in
HBase I can see the row-key of the index table and queries work as expected:
\x00\x171545413\x00 column=cf:cf:type, timestamp=1563954319353,
value=1545413
\x00\x00\x00\x01b\xB2s\xDB
@\x1B\x94\xFA\xD4\x14c\x0B
d$\x82\xAD\xE6\xB3\xDF\x06
\xC9\x07@\xB9\xAE\x00
However, for the case where the index is created asynchronously, and
then populated using the IndexTool, with the following commands:
CREATE LOCAL INDEX INDEX_TABLE ON “MyTable” (“cf”.”type”) ASYNC;
sudo -u hdfs HADOOP_CLASSPATH=`hbase classpath` hadoop jar
/opt/cloudera/parcels/CDH-5.7.1-1.cdh5.7.1.p0.11/lib/hbase/bin/../lib/hbase-client-1.2.0-cdh5.7.1.jar
org.apache.phoenix.mapreduce.index.IndexTool --data-table "MyTable"
--index-table INDEX_TABLE --output-path hdfs://nameservice1/
I get the following row-key in HBase:
\x00\x00\x00\x00\x00\x00\x column=cf:cf:type, timestamp=1563954000238,
value=1545413
00\x00\x00\x00\x00\x00\x00
\x00\x00\x00\x00\x00\x00\x
00\x00\x00\x00\x00\x00\x00
\x00\x00\x00\x00\x00\x00\x
151545413\x00\x00\x
00\x00\x01b\xB2s\xDB@\x1B\
x94\xFA\xD4\x14c\x0Bd$\x82
\xAD\xE6\xB3\xDF\x06\xC9\x
07@\xB9\xAE\x00
It is has 32 additional 0-bytes (\x00). Why is there a difference – is
one expected? What’s more, the index table in Phoenix is empty (I guess
it’s not able to read the underlying HBase index table with that key?),
so any queries that use the local index in Phoenix return no value.
Do you have any suggestions? We must use the /async /method to populate
the index table on production because of the massive amounts of data,
but if Phoenix is not able to read the index table it cannot be used for
queries.
Is it possible this issue has been fixed in a newer version?
Thanks