When using a remote OVA source, urlopen's default is to read 8192 bytes at a time, which is very slow (a simple ~1GB OVA might take tens of hours). This proposal uses http.client directly to allow setting the block size (10 MB by default, but depending on the actual environment it can be even higher). In my tests, a 10 MB block size resulted in a 80x reduction of download times from a local LAN web server, and a 100 MB block size produced a 500x reduction. Differences might not be so dramatic depending on the actual conditions, but I think we're still much better off with the new default, and anyway it's tweakable if desired.
When using a remote OVA source,
urlopen's default is to read 8192 bytes at a time, which is very slow (a simple ~1GB OVA might take tens of hours). This proposal useshttp.clientdirectly to allow setting the block size (10 MB by default, but depending on the actual environment it can be even higher). In my tests, a 10 MB block size resulted in a 80x reduction of download times from a local LAN web server, and a 100 MB block size produced a 500x reduction. Differences might not be so dramatic depending on the actual conditions, but I think we're still much better off with the new default, and anyway it's tweakable if desired.As noted in https://github.com/vmware/pyvmomi-community-samples/issues/392, for non-disk objects (like ISO, or NVRAM files)
PUTshould be used instead ofPOST. This proposal also applies the change suggested in that issue.