Stream interface built on top of azure for incrementally pushing buffers and committing them. Designed for "live" logging and bursty streams of data (which is expected to end eventually).
There are many common cases which the azure client handles much better then this library if you doing any of the following use the azure client:
The algorithm is very simple (dumb)
Due to how node streams work while we are writing the readable side will buffer its writes up to the high water mark.
This stream does not do any special internal buffering to optimize
backed up writes. What this means is if your goal is to "append" as fast
as possible without much concern for memory node 0.10 will be much
slower then 0.11 because writes are done in order and each buffer is
written sepeartely without merges of the buffers that have yet to be
written. Node 0.11 introduces
_writev which is used here to merge any
pending buffers before the write which avoids most latency issues.
var AzureStream = ;var azure = ;var blobService = azure;var azureWriter =blobService'mycontainer''myfile.txt';// any kind of node readable stream herevar nodeStream;nodeStream;azureWriter;
azure-storage module is slow to load (147ms) and takes up 19mb of
memory (as of 0.3). We don't use very many azure blob api calls so
ideally we could extract (or help the primary lib extract) the url
signing part of authentication into its own lib and then just directly
call http for our operations... The ultimate goal here is to consume
around 5mb (including https overhead) of memory and load in under 20ms.
To correctly consume the url from azure the
x-ms-version header must
be set to something like
2013-08-15 this allows open ended range
range: byte=500-). In combination with etags (and if
conditions) we can build a very fast client (even a fast polling