Repository navigation
IndexNode 0.13.5
Two read paths still failed on Nexis ghosts after 0.13.4.
- A vanished snapshot is not a missing blob. A read that went to a Nexis snapshot (the blob was caught mid-dehydrate) and got a 404 used to end as "centers.zip is gone": Nexis deletes that ephemeral snapshot a few minutes after the transition.
RestStorenow goes back to the base blob, waits for the transition to finish (about 2 minutes, with its retry backoff), hydrates when it is a ghost, and reads on (GetAsync, ranged reads including the If-Match one,HeadAsync, downloads). Only a base blob that 404s is gone. - Every zip open hydrates a ghost first. The backfill's
download.zip(duration) andtranscribe.zipreads, thedescribe.zipending backfill and the stage-zip view took the length from a HEAD (0 for a ghost) and failed with "not a zip: the blob is 0 bytes". They now all go throughZipOpen.HeadReadableAsync, likeCentersZip: a 0-byte blob taggedghostd_state=ghost(or withghostd_size> 0) is hydrated and HEADed again; a 0-byte blob with no ghost tags is still "not a zip".