Github user JamesRTaylor commented on a diff in the pull request:

    https://github.com/apache/phoenix/pull/8#discussion_r16962143
  
    --- Diff: 
phoenix-core/src/main/java/org/apache/phoenix/coprocessor/MetaDataEndpointImpl.java
 ---
    @@ -664,14 +698,161 @@ private PTable getTable(RegionScanner scanner, long 
clientTimeStamp, long tableT
               } else {
                   addColumnToTable(results, colName, famName, colKeyValues, 
columns, saltBucketNum != null);
               }
    +        }        
    +        byte[] tenIdBytes = QueryConstants.EMPTY_BYTE_ARRAY;
    +        if (tenantId != null) {
    +            tenIdBytes = tenantId.getBytes();
             }
    -
    +        byte[] schNameInBytes = QueryConstants.EMPTY_BYTE_ARRAY;
    +        if (schemaName != null) {
    +            schNameInBytes = Bytes.toBytes(schemaName.getString());
    +        }
    +        PTableStats stats = updateStats(SchemaUtil.getTableKey(tenIdBytes, 
schNameInBytes, tableNameBytes));
             return PTableImpl.makePTable(tenantId, schemaName, tableName, 
tableType, indexState, timeStamp, 
                 tableSeqNum, pkName, saltBucketNum, columns, tableType == 
INDEX ? dataTableName : null, 
                 indexes, isImmutableRows, physicalTables, defaultFamilyName, 
viewStatement, disableWAL, 
    -            multiTenant, viewType, viewIndexId, indexType);
    +            multiTenant, viewType, viewIndexId, indexType, stats);
         }
     
    +    private PTableStats updateStats(final byte[] tableNameBytes) {
    +        lock.readLock().lock();
    +        try {
    +            PTableStats stats = 
tableStatsMap.get(Bytes.toString(tableNameBytes));
    +            return stats;
    +        } finally {
    +            lock.readLock().unlock();
    +        }
    +    }
    +    
    +    private void updateStatsInternal(byte[] tableNameBytes, 
RegionCoprocessorEnvironment env)
    --- End diff --
    
    Good point. The best way to ensure other clients will re-read the metadata 
from the server is to update a timestamp on a Cell in the row. Jeffrey made a 
change like this to "force" clients to do this by doing a Put of the empty key 
value. See MetaDataEndPointImpl.updateIndexState() line 1538:
    
                        if(dataTableKey != null) {
                            // make a copy of tableMetadata
                            tableMetadata = new 
ArrayList<Mutation>(tableMetadata);
                            // insert an empty KV to trigger time stamp update 
on data table row
                            Put p = new Put(dataTableKey);
                            p.add(TABLE_FAMILY_BYTES, 
QueryConstants.EMPTY_COLUMN_BYTES, timeStamp, ByteUtil.EMPTY_BYTE_ARRAY);
                            tableMetadata.add(p);
                        }
    
    How about having your invalidate method do that instead? In this case, the 
client makes a getTable() call in MetaDataClient.updateCache() and the new 
PTable is sent across (which would include the stats).


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---

Reply via email to