…amed to qfs-access-<ver>.jar, (b) Hadoop QFS plugin is now built and tested in QFS codebase, (c) Maven2 is used to build java components (instead of Apache Ant); it also pulls hadoop dependencies. Minor modifications to build version detection script.
… synchronized. Fix potential spurious close() invocations from finalize() in the case where kfs_close raised exception in channel close method, and the file descriptor did not get invalidated / reset. Reset the channel file descriptor before invoking kfs_close. With the file descriptor numbers re-use not resetting file descriptor could potentially cause close of the valid file descriptor. Ensure that finalize methods do not invoke synchronized methods.
…ix close() status code where protocol worker effectively had no requests to process.
…cation rpc Show() method to facilitate debugging. Meta server: display "parent" chunk allocation rpc from chunk allocation rpc method Show() to facilitate debugging. Ensure that all chunk servers that chunk allocation rpc have already successfully completed the chunk allocation or version change op are still operational before declaring chunk replica lost due to (other) chunk server allocation rpc timeout, or discarding non stable chunk replica on chunk server re-connect. This is needed to handle the case where chunk server(s) go(es) down and possibly re-connect while chunk allocation is still in-flight (waiting for replies from the other remaining chunk servers).
…y content in "dir checker" and report available chunks to the chunk manager, which in turn reports the chunks that became available to the meta server. Declare the newly available chunk stale in the case when the chunk replica already exists. Handle meta server disconnect by re-queuing the available chunks that were part of the corresponding "available chunks" rpc in flight. Fix potential memory leak the meta server state machine with "no reply" rcp (presently only corrupt chunk) with non empty pending ops queue, by queuing "no reply" ops completion invocation into a temporary queue, and running the queue before the ops dispatch method returns. Make "isBeingReplicated" flag private and create and use the corresponding access methods.
…n from chunk directories checker to chunk manager. Fix chunk size calculation in "restore" with the "trimmed" chunk files.
…k server hello by immediately writing chunk information directly into the corresponding io buffers, and then move io buffers into the connection's output buffer. Separate startup chunk file validation logic into IsValidChunkFile() method. This would allow to use the same method in both chunk server startup logic and chunk directory re-use logic in DirChecker.
…ist of the columns displayed by default on chunk servers status page.
…k only in the case where the number of replicas doesn't match the file replication.
…ck has one or less recovery chunks available.
…to convey to the meta server that certail chunk have become available. The general idea is that these chunk should be used by the meta server as of last resort.