Keith Turner created ACCUMULO-2175:
--------------------------------------
Summary: Batch defining tablets in walog
Key: ACCUMULO-2175
URL: https://issues.apache.org/jira/browse/ACCUMULO-2175
Project: Accumulo
Issue Type: Improvement
Reporter: Keith Turner
Fix For: 1.7.0
If a batch of mutations comes into a tablet server AND the tablet server just
got a new walog then it will sync the walog for each tablet. Below is a sketch
of what the tablet server currently does.
{code:java}
foreach(Tablet t : tabletsInMutationBatch){
if(!tabletIsDefinedInWalog(t, currentWalog)){
defineTablet(currentWalog, t); //syncs walog
addWalogToMetadataTable(currentWalog, t); //syncronous metadata table
update
}
}
{code}
Seems like doing the following would be better. Then no matter how many
undefined tablets there are, only one walog sync would be done.
{code:java}
foreach(Tablet t : tabletsInMutationBatch){
Set<Tablet> undefined = new HashSet<Tablet>();
if(!tabletIsDefinedInWalog(t, currentWalog)){
undefined.add(t);
}
}
defineTablets(currentWalog, undefined); //syncs walog after writing all
definitions
addWalogToMetadataTable(currentWalog, undefined); //syncronous metadata table
batch write
{code}
There is not problem when all tablets in a batch update are defined in the
walog. In this case a batch update that contains multiple tablet will only sync
the log once after adding all the mutations from all tablets.
Noticed this while looking into ACCUMULO-2172
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
