Check your logs for whether your end-point coprocessor is hitting
zookeeper on every invocation to figure out the region start key.
Unfortunately (at least last time I checked), the default way of
invoking an end point coprocessor doesn't use the meta cache. You can go
through a combination of the following instead:
HRegionLocation regionLocation = retried ?
connection.relocateRegion(tableName, tableKey) :
connection.locateRegion(tableName, tableKey);
...
Then call HConnection.processExecs call, passing in the regionKeys from
above.
You can trap the error case of the region being relocated and try again
with retried = true and it'll update the meta data cache when
relocateRegion is called.
Once we made this change for Phoenix, our latencies went way down.
HTH,
James
On 03/04/2013 05:43 PM, Andrew Purtell wrote:
Do you have timing results for an Apache HBase release? Our last release
was 0.94.5.
On Tuesday, March 5, 2013, Kim Hamilton wrote:
Hi all,
I've been lurking here for a while, so thanks for all the valuable tips and
guidance you've given so far.
I'm running some experiments to understand where to use coprocessors. One
interesting scenario is computing distinct values. I ran performance tests
with two distinct value implementations: one using endpoint coprocessors,
and one using just scans (computing distinct values client side only). I
noticed that the endpoint coprocessor implementation averaged 80 ms slower
than the scan implementation. Details of that are below for anyone
interested.
To drill into the performance, I instrumented the code and ultimately
deployed a no-op endpoint coprocessor, to look at the overhead of simply
calling it. I'm measuring around 100ms for calling my empty, no-op endpoint
coprocessor.
I need to do more tests, but I believe my tests are leading me to similar
conclusions drawn here:
http://hbase-coprocessor-experiments.blogspot.com/2011/05/extending.html
I.e. if the query/scan is selective enough (I'll go out on a limb and
estimate 50-100 rows), then it's better to just perform a scan and compute
client side. Endpoint coprocessors will make sense for larger result sets
and/or scans that hit multiple regions.
Before going too far, I wanted to check if anyone in this group has
suggestions. I.e. perhaps there are just some configuration options I've
not uncovered. Does this 100ms latency sound correct?
Thanks,
Kim
*Detailed results of distinct value comparison, just FYI*
Using 0.92.1-cdh4.1.0
Scan result size ~50-100
Row size 1kb, but after filtering for only desired columns, 380 bytes
*with coprocessors*
AverageLatency(ms), 176.1353
MinLatency(ms), 42
MaxLatency(ms), 2368
95thPercentileLatency(ms), 321
99thPercentileLatency(ms), 422
*scan-only, compute distinct values client side*
AverageLatency(ms), 92.8165
MinLatency(ms), 4
MaxLatency(ms), 986
95thPercentileLatency(ms), 253
99thPercentileLatency(ms), 356
