Demai Ni created HBASE-9531:
-------------------------------
Summary: a command line (hbase shell) interface to retreive the
replication metrics and show replication lag
Key: HBASE-9531
URL: https://issues.apache.org/jira/browse/HBASE-9531
Project: HBase
Issue Type: New Feature
Components: Replication
Reporter: Demai Ni
Assignee: Demai Ni
This jira is to provide a command line (hbase shell) interface to retreive the
replication metrics info such as:ageOfLastShippedOp, timeStampsOfLastShippedOp,
sizeOfLogQueue ageOfLastAppliedOp, and timeStampsOfLastAppliedOp. And also to
provide a point of time info of the lag of replication(source only)
Understand that hbase is using Hadoop
metrics(http://hbase.apache.org/metrics.html), which is a common way to monitor
metric info. This Jira is to serve as a light-weight client interface,
comparing to a completed(certainly better, but heavier)GUI monitoring package.
I made the code works on 0.94.9 now, and like to use this jira to get opinions
about whether the feature is valuable to other users/workshop. If so, I will
build a trunk patch.
All inputs are greatly appreciated. Thank you!
The overall design is to reuse the existing logic which supports hbase shell
command 'status', and invent a new module, called ReplicationLoad. In
HRegionServer.buildServerLoad() , use the local replication service objects to
get their loads which could be wrapped in a ReplicationLoad object and then
simply pass it to the ServerLoad. In ReplicationSourceMetrics and
ReplicationSinkMetrics, a few getters and setters will be created, and ask
Replication to build a "ReplicationLoad". (many thanks to Jean-Daniel for his
kindly suggestions through dev email list)
the replication lag will be calculated for source only, and use this formula:
{code:title=Replication lag|borderStyle=solid}
if sizeOfLogQueue != 0 then max(ageOfLastShippedOp, (current time -
timeStampsOfLastShippedOp)) //err on the large side
else if (current time - timeStampsOfLastShippedOp) < 2*
ageOfLastShippedOp then lag = ageOfLastShippedOp // last shipped happen
recently
else lag = 0 // last shipped may happens last night, so NO real
lag although ageOfLastShippedOp is non-zero
{code}
External will look something like:
{code:title=status 'replication'|borderStyle=solid}
hbase(main):001:0> status 'replication'
version 0.94.9
3 live servers
hdtest017.svl.ibm.com:
SOURCE:PeerID=1, ageOfLastShippedOp=14, sizeOfLogQueue=0,
timeStampsOfLastShippedOp=Wed Sep 04 14:49:48 PDT 2013
SINK :AgeOfLastAppliedOp=0, TimeStampsOfLastAppliedOp=Wed Sep 04
14:48:48 PDT 2013
hdtest018.svl.ibm.com:
SOURCE:PeerID=1, ageOfLastShippedOp=0, sizeOfLogQueue=0,
timeStampsOfLastShippedOp=Wed Sep 04 14:48:48 PDT 2013
SINK :AgeOfLastAppliedOp=14, TimeStampsOfLastAppliedOp=Wed Sep 04
14:50:59 PDT 2013
hdtest015.svl.ibm.com:
SOURCE:PeerID=1, ageOfLastShippedOp=0, sizeOfLogQueue=0,
timeStampsOfLastShippedOp=Wed Sep 04 14:48:48 PDT 2013
SINK :AgeOfLastAppliedOp=0, TimeStampsOfLastAppliedOp=Wed Sep 04
14:48:48 PDT 2013
hbase(main):002:0> status 'replication','source'
version 0.94.9
3 live servers
hdtest017.svl.ibm.com:
SOURCE:PeerID=1, ageOfLastShippedOp=14, sizeOfLogQueue=0,
timeStampsOfLastShippedOp=Wed Sep 04 14:49:48 PDT 2013
hdtest018.svl.ibm.com:
SOURCE:PeerID=1, ageOfLastShippedOp=0, sizeOfLogQueue=0,
timeStampsOfLastShippedOp=Wed Sep 04 14:48:48 PDT 2013
hdtest015.svl.ibm.com:
SOURCE:PeerID=1, ageOfLastShippedOp=0, sizeOfLogQueue=0,
timeStampsOfLastShippedOp=Wed Sep 04 14:48:48 PDT 2013
hbase(main):003:0> status 'replication','sink'
version 0.94.9
3 live servers
hdtest017.svl.ibm.com:
SINK :AgeOfLastAppliedOp=0, TimeStampsOfLastAppliedOp=Wed Sep 04
14:48:48 PDT 2013
hdtest018.svl.ibm.com:
SINK :AgeOfLastAppliedOp=14, TimeStampsOfLastAppliedOp=Wed Sep 04
14:50:59 PDT 2013
hdtest015.svl.ibm.com:
SINK :AgeOfLastAppliedOp=0, TimeStampsOfLastAppliedOp=Wed Sep 04
14:48:48 PDT 2013
hbase(main):003:0> status 'replication','lag'
version 0.94.9
3 live servers
hdtest017.svl.ibm.com: lag = 0
hdtest018.svl.ibm.com: lag = 14
hdtest015.svl.ibm.com: lag = 0
{code}
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
