[
https://issues.apache.org/jira/browse/IGNITE-26004?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Vladislav Pyatkov updated IGNITE-26004:
---------------------------------------
Description:
h3. Motivation
We do not know anything about messages that are processed through the network
layer. So we cannot observe the image's quantity and size. It restrains
investigation because we might not notice the network overload.
Here is a histogram that I got manually from the server-server communication
channel.
{noformat}
Server-server communication
org.apache.ignite.internal.network.message.InvokeRequestImpl:TxCleanupMessageImpl=139570,
org.apache.ignite.internal.network.message.InvokeRequestImpl:AppendEntriesRequestImpl=1371190,
org.apache.ignite.internal.network.message.InvokeRequestImpl:CoalescedHeartbeatRequestImpl=99,
org.apache.ignite.internal.network.message.InvokeResponseImpl:TxCleanupMessageResponseImpl=115714,
org.apache.ignite.internal.network.message.InvokeResponseImpl:CoalescedHeartbeatResponseImpl=100,
org.apache.ignite.internal.tx.message.TxCleanupMessageResponseImpl=115715,
org.apache.ignite.internal.network.message.InvokeResponseImpl:AppendEntriesResponseImpl=1094870
{noformat}
and server client
{noformat}
Client-server communication
ClientOp.TX_COMMIT=68919,
ClientOp.TUPLE_UPSERT=344137
{noformat}
Moreover, it would be appropriate to make a histogram across message size.
h3. Implementation notes
We should take into account that we have several network threads, and each
handles a specific load.
h3. Definition of done
Add the histogram metrics.
was:
h3. Motivation
We do not know anything about messages that are processed through the network
layer. So we cannot observe the image's quantity and size. It restrains
investigation because we might not notice the network overload.
Here is a histogram that I got manually from the server-server communication
channel.
{noformat}
Server-server communication
org.apache.ignite.internal.network.message.InvokeRequestImpl:TxCleanupMessageImpl=139570,
org.apache.ignite.internal.network.message.InvokeRequestImpl:AppendEntriesRequestImpl=1371190,
org.apache.ignite.internal.network.message.InvokeRequestImpl:CoalescedHeartbeatRequestImpl=99,
org.apache.ignite.internal.network.message.InvokeResponseImpl:TxCleanupMessageResponseImpl=115714,
org.apache.ignite.internal.network.message.InvokeResponseImpl:CoalescedHeartbeatResponseImpl=100,
org.apache.ignite.internal.tx.message.TxCleanupMessageResponseImpl=115715,
org.apache.ignite.internal.network.message.InvokeResponseImpl:AppendEntriesResponseImpl=1094870
{noformat}
and server client
{noformat}
Client-server communication
ClientOp.TX_COMMIT=68919,
ClientOp.TUPLE_UPSERT=344137
{noformat}
Moreover, it would be appropriate to make a histogram across message size.
> Network messages histogramm
> ---------------------------
>
> Key: IGNITE-26004
> URL: https://issues.apache.org/jira/browse/IGNITE-26004
> Project: Ignite
> Issue Type: Improvement
> Reporter: Vladislav Pyatkov
> Priority: Major
> Labels: ignite-3
>
> h3. Motivation
> We do not know anything about messages that are processed through the network
> layer. So we cannot observe the image's quantity and size. It restrains
> investigation because we might not notice the network overload.
> Here is a histogram that I got manually from the server-server communication
> channel.
> {noformat}
> Server-server communication
> org.apache.ignite.internal.network.message.InvokeRequestImpl:TxCleanupMessageImpl=139570,
> org.apache.ignite.internal.network.message.InvokeRequestImpl:AppendEntriesRequestImpl=1371190,
> org.apache.ignite.internal.network.message.InvokeRequestImpl:CoalescedHeartbeatRequestImpl=99,
> org.apache.ignite.internal.network.message.InvokeResponseImpl:TxCleanupMessageResponseImpl=115714,
> org.apache.ignite.internal.network.message.InvokeResponseImpl:CoalescedHeartbeatResponseImpl=100,
> org.apache.ignite.internal.tx.message.TxCleanupMessageResponseImpl=115715,
> org.apache.ignite.internal.network.message.InvokeResponseImpl:AppendEntriesResponseImpl=1094870
> {noformat}
> and server client
> {noformat}
> Client-server communication
> ClientOp.TX_COMMIT=68919,
> ClientOp.TUPLE_UPSERT=344137
> {noformat}
> Moreover, it would be appropriate to make a histogram across message size.
> h3. Implementation notes
> We should take into account that we have several network threads, and each
> handles a specific load.
> h3. Definition of done
> Add the histogram metrics.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
