JobManager

LibraryCacheManager

BlobServer

local (rw)

 

<path>/<jobId>/<BlobKey>

 

HA (rw)

 

<path>/<jobId>/<BlobKey>

copy

recover

request/release jar,
refCount±1

TaskManager

LibraryCacheManager

BlobCache

local (rw)

 

<path>/<jobId>/<BlobKey>

HA (r)

 

<path>/<jobId>/<BlobKey>

copy

request/release jar,
refCount±1

BlobClient

download/upload BLOBs

download

RpcService

upload message (refCount+x);

download message (refCount-1)

RpcService

download message (refCount+1);
deserialise message (refCount-1)

ack message, refCount- 1

upload message;

refCount+x

WebUI

download log (refCount+1);
release log (refCount-1)

request log upload

* refCount
* cleanup

* refCount
* cleanup

upload log

Job Submission Client

BlobClient

upload BLOBs