supporting mtime (and mode, i.e., chmod) via s3fs

[rri...@gmail.com]

just a continuation of http://rrizun.blogspot.com/2007/10/s3fs-rsync-and-mtime.html

supporting mtime (and mode, i.e., chmod) via s3fs so that tools like
rsync would work

right now s3fs does not support changing mtime (which is what rsync
wants to do)

1. use "standard" amazon s3 x-amz- headers to store "mtime" and "mode"

pros: would be "standard" (i.e., doing it the way amazon would
intend); would not have to "pollute" the bucket with custom meta
objects

cons: would be extremely slow: would effectively *triple* the amount
of data transfered for, say, a chmod cmd (because s3fs would have to
download the object and then re-upload it with new/updated meta)

2. use small "proprietary" meta objects in s3 for each object to store
name/value pairs.

pros: performance would be pretty much as fast as could be

cons: would pollute the bucket with custom meta objects (you would of
course only see this with 3rd party s3 clients; s3fs would of course
not show 'em in dir listings, etc...)

I'm leaning toward #2.

[Roberto Saccon]

I think there is no right or wrong here. Maybe supporting both
variants, the user chooses the preferred one by setting he "advanced"
options.

On Oct 17, 3:40 pm, "rri...@gmail.com" <rri...@gmail.com> wrote:
> just a continuation ofhttp://rrizun.blogspot.com/2007/10/s3fs-rsync-and-mtime.html

[Randy Rizun]

FYI I've posted a new version of s3fs (svn version 43) that has
support for mode and mtime (via "x-amz-meta-mode" and
"x-amz-meta-mtime" custom meta headers)

rsync works now! first/initial rsync is slow(er), but, it does work...
but then, incremental rsyncs are pretty much as fast as can be