Improve logging in channel pipeline #431
Conversation
lutovich
force-pushed
the
1.5-improve-logging
branch
2 times, most recently
from
d3bb70a to
12896b9
Compare
| if ( log.isDebugEnabled() ) | ||
| { | ||
| log.debug( "Received RECORD message with metadata %s", Arrays.toString( fields ) ); | ||
| log.debug( "Received RECORD message with metadata %s for handler %s", Arrays.toString( fields ), handler ); |
There was a problem hiding this comment.
Print all incoming messages as in LoggingResponseHandler in 1.4
| readMessageBoundary = ((ByteBuf) msg).readableBytes() == 0; | ||
| readMessageBoundary = chunkBuf.readableBytes() == 0; | ||
|
|
||
| if ( log.isTraceEnabled() ) |
There was a problem hiding this comment.
Consider logging chunk header and message boundary so that we can reconstruct messages from hex when debugging.
| ClusterComposition cluster = response.clusterComposition(); | ||
| logger.info( "Got cluster composition %s", cluster ); | ||
| return cluster; | ||
| return response.clusterComposition(); |
There was a problem hiding this comment.
Consider logging records received from routing procedure on debug level
There was a problem hiding this comment.
If debug level is enabled, then the routing table will be printed out in the record message coming from getRoutingTable procedure. Therefore there might be no need to log again in debug level.
Instead we might want a logger at INFO level when there is a new routing table if we only have INFO level on.
|
@zhenlineo comments should now be fixed |
| @@ -77,7 +92,7 @@ protected void encode( ChannelHandlerContext ctx, Message msg, List<Object> out | |||
|
|
|||
| if ( log.isTraceEnabled() ) | |||
| { | |||
| log.trace( "Message %s encoded as\n%s\n", msg, prettyHexDump( messageBuf ) ); | |||
| log.trace( "C: %s encoded as\n%s", msg, prettyHexDump( messageBuf ) ); | |||
There was a problem hiding this comment.
Shall we change this to "C:\n%s" so that
- it is the same as
log.trace( "S:\n%s", prettyHexDump( msg ) ); - to avoid printing message twice as it already printed out in debug.
| ClusterComposition cluster = response.clusterComposition(); | ||
| logger.info( "Got cluster composition %s", cluster ); | ||
| return cluster; | ||
| return response.clusterComposition(); |
There was a problem hiding this comment.
If debug level is enabled, then the routing table will be printed out in the record message coming from getRoutingTable procedure. Therefore there might be no need to log again in debug level.
Instead we might want a logger at INFO level when there is a new routing table if we only have INFO level on.
| log.info( "Got getServers response: %s", records ); | ||
| if ( log.isDebugEnabled() ) | ||
| { | ||
| log.debug( "Received response from %s procedure: %s", invokedProcedureString( response ), records ); |
There was a problem hiding this comment.
sugget to change to INFO
Made all handlers in the channel pipeline prefix their log messages with channel info. Added tests and fixed issue where code would hang if channel gets disconnected during bolt handshake. Improved layout of java packages.
* `C:` and `S:` message prefixes as in other driver verisons * removed logging of chunks without headers * added trace logging of full chunks and message boundary * log raw routing procedure response on debug level
Usages replaced with `io.netty.buffer.ByteBufUtil`.
* removed double printing of outgoing message object * changed logging of routing procedure response to info level
lutovich
force-pushed
the
1.5-improve-logging
branch
from
9c4831e to
05f5915
Compare
Made all handlers in the channel pipeline prefix their log messages with channel info. Added tests and fixed issue where code would hang if channel gets disconnected during bolt handshake. Improved layout of java packages.