cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Paulo Motta (JIRA)" <>
Subject [jira] [Commented] (CASSANDRA-11303) New inbound throughput parameters for streaming
Date Fri, 03 Jun 2016 14:06:59 GMT


Paulo Motta commented on CASSANDRA-11303:

Thanks for the update [~skonno]! This looks good and we're getting close, just a few minor
* Rename StorageService.(get/set)(Inbound/Outbound)StreamThroughputMbPerSec to be consistent
with DatabaseDescriptor.(get/set)StreamThroughput(Inbound/Outbound)MegabitsPerSec (even though
it's a bit uglier, it's been there first so we should probably keep it)
* Add stream_throughput_inbound_megabits_per_sec to inter_dc_stream_throughput* property description
on cassandra.yaml
* Add deprecation javadoc to deprecated StorageServiceMBean methods (see forceRepairAsync
for example)
* remove additional spaces from  throw new IOException("CF   "    + cfId + " was dropped during
* It seems your IDE removed static qualifiers on StreamReader constants and static methods
* Keep old methods on NodeProbe (add deprecation flag) in case they're used externally
* Right now we're only setting the throughput when the rate limiter is created, but we must
also update it on StorageService.set*StreamThroughput methods, so the new rate is valid for
ongoing streams. We can probably simplify this to set the rate only on static initialization
and when set*StreamThroughput methods are called for dynamic updating.
** After this we can probably get rid of specific Inbound/Outbound and have just a single
StreamRateLimiter that is created with a limiter/interDcLimiter by the factory method

I tested the patch locally with default settings (25MiB/s) vs throttling inbound at 5MiB/s
and worked like a charm (attached bmon output). I will test updating it dynamically during
ongoing stream with the new version.

I also submitted dtests and unit tests on CI:


> New inbound throughput parameters for streaming
> -----------------------------------------------
>                 Key: CASSANDRA-11303
>                 URL:
>             Project: Cassandra
>          Issue Type: New Feature
>          Components: Configuration
>            Reporter: Satoshi Konno
>            Priority: Minor
>         Attachments: 11303_inbound_limit_debug_20160419.log, 11303_inbound_nolimit_debug_20160419.log,
11303_inbound_patch_for_trunk_20160419.diff, 11303_inbound_patch_for_trunk_20160525.diff,
200vs40inboundstreamthroughput.png, cassandra_inbound_stream.diff
> Hi,
> To specify stream throughputs of a node more clearly, I would like to add the following
new inbound parameters like existing outbound parameters in the cassandra.yaml.
> - stream_throughput_inbound_megabits_per_sec
> - inter_dc_stream_throughput_outbound_megabits_per_sec  
> We use only the existing outbound parameters now, but it is difficult to control the
total throughputs of a node. In our production network, some critical alerts occurs when
a node exceed the specified total throughput which is the sum of the input and output throughputs.
> In our operation of Cassandra, the alerts occurs during the bootstrap or repair processing
when a new node is added. In the worst case, we have to stop the operation of the exceed node.
> I have attached the patch under consideration. I would like to add a new limiter class,
StreamInboundRateLimiter, and use the limiter class in StreamDeserializer class. I use Row::dataSize(
)to get the input throughput in StreamDeserializer::newPartition(), but I am not sure whether
the dataSize() returns the correct data size.
> Can someone please tell me how to do it ?

This message was sent by Atlassian JIRA

View raw message