CreateObject is a single HTTP transaction. Should it fail, you must re-send the entire request. If you have large objects and require a "resume" behavior, then you should buffer the data and send CreateObject followed by multiple UpdateObject (append) calls. If an append operation fails, you can resume from the last successful append.
If you're using checksums the process is more complex. In that case, you'll have to HEAD the object to find out from what offset to resume. Then you'll have to issue a new append request from that offset and send the remainder of the buffer.
You should also upload multiple objects in parallel to increase throughput, especially with high latency.
chris_arnett
110 Posts
0
July 23rd, 2014 09:00
Lachezar,
CreateObject is a single HTTP transaction. Should it fail, you must re-send the entire request. If you have large objects and require a "resume" behavior, then you should buffer the data and send CreateObject followed by multiple UpdateObject (append) calls. If an append operation fails, you can resume from the last successful append.
If you're using checksums the process is more complex. In that case, you'll have to HEAD the object to find out from what offset to resume. Then you'll have to issue a new append request from that offset and send the remainder of the buffer.
You should also upload multiple objects in parallel to increase throughput, especially with high latency.