collector. This will allow greater insight into the generations.
Now the client will emmit the stats for each collector at the end of
run along with the totals.
the JVM spent in GC. It now prints the time and percent of time in
the final output.
Also fix cases where the garbage collector implementation returns
-1 if the count or time isn't implemented.
This changeset refactors the scan operation slightly, using a List
with a predefined size (since we know it), as well as using the
document ID directly for the response rather than doing full blown
JSON decoding which is not needed.
This changeset picks better default settings for the environment
to make it less error-prone in cloud setups and in turn leads to
easier usability.
The changes made are:
- queryEndpoints set from 5 to 1 by default since this is technically
a leftover from earlier implementations. We now have a Bucket per
Thread and since YCSB is sync anyways we only ever need one. Leaving
it as an option since if we get async ops in the future we can utilize
those.
- socketConnectTimeout from 1s to 10s, especially in cloud environments
socket timeouts are annoying and its better if we have higher defaults.
- connectTimeout changed from 5s to 30s since the socket connect timeout
is higher we also need to adjust this one.
- kvTimeout changed from 2.5s to 10s so that under throughput saturation
testing you don't get flooded with timeouts potentially. Note that it
also makes for nicer output if you get "higher latencies" as part of
the regular output and not in the error section.
[riak] Moved the utility function createResultHashMap() to the utility class RiakUtils.java.
[riak] Added riak.strong_consistent_scans_bucket_type property to riak.properties file to set the fake bucket-type name.
[riak] Updated README.md with a guide on how to allow strong-consistent scan transactions.
This commit modifies the Elasticsearch client to return Status.NOT_FOUND
when an operation fails because the document requested does not
exist. This modifies the previous behavior which would return
Status.ERROR in this case.
This commit also moves the statements that would return Status.ERROR
into the corresponding catch blocks so that it is clear that
Status.ERROR is only being returned on error paths.
when the status thread is enabled. This tracks the GC count, thread count,
memory used and average system load (if implemented) over time as well as
the max and min for each measurement at the end.
Also, every run will now measure the total number of garbage collections
from all pools.
Note the Guava classes in the test directory for AsyncHBase. These are there
as AsyncHBase uses Guava >= 18 while HBase and the mini cluster use Guava 12.
This will be used for #653 for databases that require keys or other values to use a limited character set. This generator will create a sequence of strings using a defined list of UTF code points from which to pull.
Chris Larsen
Chrisjan Matser
Camilo Aguilar
Chris Larsen
Sean Busbey
Kevin Risden
Kevin Risden
Jaemyoun
Anthony Baker
Sean Busbey
Michael Nitschinger
danielpoltx
nygard89
Jean-Daniel Cryans
Jason Tedor
yuyanting