Details
-
Bug
-
Status: Closed
-
Major
-
Resolution: Duplicate
-
0.98.8
-
None
-
None
-
None
Description
when put/get from hbase, if we meet a temporary dns failure causes resolve RS's host, the error will never recovered. put/get will failed with UnknownHostException forever.
I checked the code, and the reason maybe:
1. when RegionServerCallable or MultiServerCallable prepare(), it gets a ClientService.BlockingInterface stub from Hconnection
2. In HConnectionImplementation::getClient, it caches the stub with a BlockingRpcChannelImplementation
3. In BlockingRpcChannelImplementation(),
this.isa = new InetSocketAddress(sn.getHostname(), sn.getPort()); If we meet a temporary dns failure then the "address" in isa will be null.
4. then we launch the real rpc call, the following stack is:
Caused by: java.net.UnknownHostException: unknown host: xxx.host2
at org.apache.hadoop.hbase.ipc.RpcClient$Connection.<init>(RpcClient.java:385)
at org.apache.hadoop.hbase.ipc.RpcClient.createConnection(RpcClient.java:351)
at org.apache.hadoop.hbase.ipc.RpcClient.getConnection(RpcClient.java:1523)
at org.apache.hadoop.hbase.ipc.RpcClient.call(RpcClient.java:1435)
at org.apache.hadoop.hbase.ipc.RpcClient.callBlockingMethod(RpcClient.java:1654)
at org.apache.hadoop.hbase.ipc.RpcClient$BlockingRpcChannelImplementation.callBlockingMethod(RpcClient.java:1712)
Besides, i noticed there is a protection in RpcClient:
if (remoteId.getAddress().isUnresolved())
shouldn't we do something when this situation occurred?
Attachments
Attachments
Issue Links
- duplicates
-
HBASE-15856 Cached Connection instances can wind up with addresses never resolved
- Resolved
- links to