Right now, in the finishLocalSampling method of the ColumnFamilyStore
we return the size of the list of returned partitions. Instead, we should
be propagating the actual cardinality of the sampled set.
Let's just read the read_cardinality and write_cardinality properties
of the scylla's REST API response.
Fixes#148
Fixes#134
Refs #135
Replaces previous refresh calls with ones bound to registration
check objects, which provides some sync between threads doing
refresh, and reduced redundant calls.
Also adds repeated reaping of dead objects, i.e. every 5 minutes
we try to remove dead CF:s (not adding new ones), to reduce
idle footprint.
Cassandra 3.0 version of the JMX added a parameter that allows accepting
the parameter as hex.
This breaks the current implementation with a NoSuchMethodException.
This patch adds the missing implementation.
For a full support, a follow up patch in Scylla is needed, but for the
current functionality it would work.
After this patch usage example:
nodetool getsstables keyspace1 standard1 39303138374b4d343830
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
When calling nodetool toppartitions with size limit, finishLocalSampling
should respect that and limit the number of the results.
Example:
$ nodetool toppartitions -k 2 keyspace1 standard1 20
WRITES Sampler:
Cardinality: ~2 (256 capacity)
Top 2 partitions:
Partition Count +/-
38333032394d4f4d5030 4 3
4e353937383137503330 4 3
READS Sampler:
Cardinality: ~2 (256 capacity)
Top 2 partitions:
Nothing recorded during sampling period...
Fixes#66
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
This patch adds the implementation for begin and finish local sampling
of a column family.
There is a difference in the implementation of Cassandra API and Scylla.
In Cassandra and the JMX an external source start and stop the sampling.
In Scylla, a single API call start the sampling and return with the
result. In Scylla the API call always return sampling of the read and of
the writes.
To bridge the difference, the begin sampling command will use a Future
when calling the API. The finish method will wait for the future to end.
Because of the different implementation, it is possible that two
consecutive calls will be made to start sampling one for the read and
one for the write, similarly, two calls will be made to finish for read
and write.
The implementation would ignore the second call to start and will
store the result, so the second call to finish will be served from the
stored result.
Note, that the use of future is only for safety, the way we expect it to
work, the caller to the begin sampling will sleep anyhow while waiting
for the result.
To avoid breaking the MBean compatibility we piggyback the duration on
top of the sampler string.
If no duration is given, a default duration will be taken, this is also
just as a precaution, we will modify the nodetool implementation to
pass that information.
There is a known issue with cardinality, that will need to be addressed.
Also we return a value in the raw column to match what Cassandra JMX
returns, but it's a duplication of the partition key.
See scylladb/scylla#2811
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <20190128143505.5241-1-amnon@scylladb.com>
JmxMBeanServer is a concrete implementation of a MBeanServer.
We want to use it directly because we need to bypass calls to
JmxMBeanServer.registerMBean and JmxMBeanServer.unregisterMBean.
They take ObjectName as parameter, copy it using
ObjectName.getInstance(ObjectName) and pass it to registerMBean
and unregisterMBean of JmxMBeanServer.getMBeanServerInterceptor().
We want to avoid this copy and pass the ObjectName argument directly
to JmxMBeanServer.getMBeanServerInterceptor().
To do that this patch:
1. changes all MBeanServer variables to JmxMBeanServer
2. creates JmxMBeanServer in APIBuilder making sure accessing
interceptors is allowed
3. makes sure that JmxMBeanServer.getMBeanServerInterceptor().registerMBean
is called wherever JmxMBeanServer.registerMBean was called
4. makes sure that JmxMBeanServer.getMBeanServerInterceptor().unregisterMBean
is called whenever JmxMBeanServer.unregisterMBean was called
Next patch will use different ObjectName implementation that will
use less memory and this patch is crucial because without it every ObjectName
is transformed with ObjectName.getInstance which turns the object into
a regular ObjectName.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
This expose the ColumnFamilyStore registration via static method.
It would allow an external object (ie. MBeanServer) to update the
registration on demand.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
This patch fix the exception handling for connection problem, instead of
ProcessingException it now expect IllegalStateException.
The rest of the functionality remains the same.
Fixes#26
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <1458602355-23601-1-git-send-email-amnon@scylladb.com>
This patch clean the redundant output the jmx proxy creates.
It set the trace level of the called method to finest and remove some
println leftovers.
Fixes#22
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
When getSSTableCountPerLevel is called and the system is not using level
compaction the expected return is null and not an empty array.
This fix (scylla-jmx)
Fixes#11
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
The column family store API was changed so it would have a single API to
return the snapshot size.
This changes the JMX to use the same API regardless if it is called from
the ColumnFamilyMetrics or from ColumnFamilyStore.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
The API returns errors with an HTTP code 400, 404 or 500 depending on
the cause with a json object that contains the failure reason. The
error message should be populate to the JMX calling client, translated
to the appropriate exception.
This patch adds the ability to detect API failure and throw a runtime
exception with the returned message.
It is up to the calling method what to do with the exception, if it
would do nothing, the calling client would get a RuntimeException,
depends on origin MBean definition, the caller can catch the exception
and throw a specific kind.
Note that any exception that is thrown must be known to the JMX client
or it will not be able to process it.
As a first step, this patch replaces the jersey client to the newer
version under glassfish, which has an easy way of getting a Reply
object, and check its status before returning the results.
The only difference in the method that uses the APIClient is the use of
MultivaluedHashMap.
The following MBean implementation where changed
ColumnFamily
CompactionManager
Gossiper
EndpointSnitchInfo
CacheService
StorageProxy
StorageService
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
This patch uses the estimated latency that was added to the column
family metrics to get the recent and estimated latency.
It follows the same logic as origin does to call the logic in metrics.
The following method implementation will be added:
getMemtableColumnsCount
getRecentSSTablesPerReadHistogram
getSSTablesPerReadHistogram
getLifetimeReadLatencyHistogramMicros
getRecentReadLatencyHistogramMicros
getRecentReadLatencyMicros
getLifetimeWriteLatencyHistogramMicros
getRecentWriteLatencyHistogramMicros
getRecentWriteLatencyMicros
getRangeCount
getTotalRangeLatencyMicros
getLifetimeRangeLatencyHistogramMicros
getRecentRangeLatencyHistogramMicros
getRecentRangeLatencyMicros
Signed-off-by: Amnon Heiman <amnon@cloudius-systems.com>
The logic of that timer, should be that after some defined time from the
previous request a new one will be sent, the actuall rate is
meaningless and only cause delyed request to be sent in a higher
frequency after the delayed reponse was returned.
Signed-off-by: Amnon Heiman <amnon@cloudius-systems.com>
Some of the depricated method in column family are replaced by counters
in Origin. Still, some tools, like nodetool uses the depricated API.
This adds an implementation to some of the depricated method that is
based on the column family counters similiar to origin.
Signed-off-by: Amnon Heiman <amnon@cloudius-systems.com>
This uses the column family timer to check that the API is still
available.
After MAX_RETRY (currently set to 30s) the proxy will shutdown.
This address issue #2
Signed-off-by: Amnon Heiman <amnon@cloudius-systems.com>
The getEstimatedRowSizeHistogram is based on the column_family.json that
defined in the API.
The implementation in the MBean and in the metrics is now works with the
get histogram as long array method in the client.
Signed-off-by: Amnon Heiman <amnon@cloudius-systems.com>
This patch changes the MBeans to 2.1 version, with the changes an empty
stub implementation was added to the implementation so the code can
compile.
Signed-off-by: Amnon Heiman <amnon@cloudius-systems.com>
This patch cleans the file formatting of ColumnFamilyStore, CommitLog,
StorageProxy and StorageService.
Those files will be modified when moving to 2.1
Signed-off-by: Amnon Heiman <amnon@cloudius-systems.com>
The ColumnFamilyMetrics is used to report statistic of a column family.
This adds an adaptation for the ColumnFamilyMetrics that works with the
API that define in column_family.json
It adds the ColumnFamilyMetrics to the ColumnFamilyStore and expose the
keyspace name, so the ColumnFamilyMetrics would have access to it.
Signed-off-by: Amnon Heiman <amnon@cloudius-systems.com>
This patch adds the implementation for the ColumnFamilyStore by adding
calls to the relevent API that define in column_family.json
This patch also do some clean up, by removing leading underscore from
parameter names and trailing white spaces.
Signed-off-by: Amnon Heiman <amnon@cloudius-systems.com>
This adds the ColumnFamilyStoreMBean with most of its method stub. The
only current functionality is returnning the column family name and the
registration of the StoreMBean.
Signed-off-by: Amnon Heiman <amnon@cloudius-systems.com>
