* implement client options, support multiple namenodes with failover
* Fix an alignment bug causing us to send very small packets
The faulty math in block_write_stream was causing us to always send
512 byte packets.
* rename host to namenodeHost
* document standby retry behaviour
* include detail in namenode error
* deprecate WrapNamenodeConnection
* replace goto behaviour with for loop
* fallback to error from last failure
* travis: bump go version to 1.x
* Regenerate proto files
(this should fix travis issues, hopefully)
* travis: touch *.pb.go before running make
For some reason, make on travis recently started deciding that the
*.pb.go files were older than the *.proto files, ever though they
originate from the same `git checkout`. The easiest way to fix this
is simply to touch them right before we run make, so that they look
newer.
* travis: attempt to fix hadoop tarball caching
* Fix comment for CopyToRemote
Fixes#89, #90
* Deprecate NewForUser and NewForConnection
* Add the ability for 'put -' to read from stdin
Fixes#44, closes#98
* Convert from govendor to dep
* Update dependencies
* Tweak put tests
Add a very basic test, and put everything in the same directory so it
gets cleaned up properly.
* Improve conf loading and surrounding tests
- Fix a bug where configuration wasn't loaded from specified paths
- Only search one path at a time, rather than deduping from all
- Correctly handle fs.default.name
- Improve test coverage and documentation
* Fix the merge from upstream
* more fixes to merge master
* more fixes
- Fix a bug where configuration wasn't loaded from specified paths
- Only search one path at a time, rather than deduping from all
- Correctly handle fs.default.name
- Improve test coverage and documentation
For some reason, make on travis recently started deciding that the
*.pb.go files were older than the *.proto files, ever though they
originate from the same `git checkout`. The easiest way to fix this
is simply to touch them right before we run make, so that they look
newer.
* implement client options, support multiple namenodes with failover
* Fix an alignment bug causing us to send very small packets
The faulty math in block_write_stream was causing us to always send
512 byte packets.
* rename host to namenodeHost
* document standby retry behaviour
* include detail in namenode error
* deprecate WrapNamenodeConnection
* replace goto behaviour with for loop
* fallback to error from last failure
The previous implementation considered existing blocks immutable, which is not
correct.
This commit attempts to unfinalize the last existing block during an append
operation so that it may be written to again. This is done by submitting the
correct stage `hdfs.OpWriteBlockProto_PIPELINE_SETUP_APPEND` during the
initial OpWriteBlock operation. Additionally, `blockWriteStream`s are now
always initialized with an offset, which will be > 0 in case of append
operation. Finally, the GenerationStamp field is incremented at the beginning
of the append operation, which is an incrementing integer that records the
version of a particular block.
microsoft-github-policy-service[bot]
Junjie Qian
Colin Marc
Shastick
Richard Jones
Wang Dafeng
Junjie Qian
Shane Kumpf
adam
Marko Radmilac
Tyler Sommer
Marko Radmilac