On Thu, 11 Feb 2016, Jens Rosenboom wrote: > Does the digest of an object also include its metadata? Or is possibly > deep-scrubbing a pool full of objects of size 0 completely pointless? It includes the omap (key/value) data stored in an object, which isn't reflected in the object size. > I'm asking because we are seeing lot of blocked requests during > deep-scrubs for one particular PG in our index pool for a cluster > containing around 200M objects. The scrub also takes a couple of > minutes instead of a couple of seconds that we see for all the other > PGs in that pool, but I haven't yet found out what makes this > particular PG so special. You probably have a large RGW bucket whose index object happens to live in that PG. The scrub iterates over its keys and values to compare checksums, which can take a while. There's a bucket sharding capability that should be used when you know a bucket will grow very large. Big index objects can cause other annoying behavior besides just scrub. For example, recovery can be very slow and block index operations if one of the replicas drops out of the PG and needs to be brought back in sync. Unfortunately there's no [automatic] resharding capability yet in RGW... that will need to be tackled sooner or later (hopefully sooner!). sage -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
