[tahoe-dev] prodnet private-storage users: you need to do deep-add-lease soon

Brian Warner warner-tahoe at allmydata.com
Tue Mar 10 18:23:31 PDT 2009


We're getting ready to do a garbage-collection pass on the
allmydata.com prodnet. This is a mark-and-sweep sort of operation. The
"mark" tool runs through all the valid accounts, does a deep-traverse
of each rootcap, and does a "check and add-lease" on each
file/directory. The "sweep" tool will run on the storage servers, and
will delete any share that hasn't been marked recently enough.

A few folks have data in the production grid which is not reachable
from an account rootcap. To keep your data from being GC'ed, you need
check out a trunk tahoe tree, start up a local tahoe node, and then run
the following command on each of your private rootcaps:

 tahoe deep-check --add-lease ROOTCAP

If you'd like to see what's going on, add a --raw argument before or
after the --add-lease: it will then print out one (long) line of JSON
for each file or directory it encounters (including check results). The
last line will be accumulated statistics (number of files, number of
directories, file-size histogram, etc).

This process seems to run at about 1 or 2 checks-per-second (depending
upon latency), so a thousand files+directories might take 20 minutes or
so. In the long run, you'll need to do this about once a month, but
we'll make an announcement before we establish a repeating process.

Note that if your data *can* be reached through an allmydata.com
rootcap, then you don't need to do this: our own mark-process (which is
already running against a list of currently-valid accounts) will update
the leases. There are only a small handful of folks (all of whom should
know who they are) who need to do this.

cheers,
 -Brian


More information about the tahoe-dev mailing list