stop re-reading if increasing minrecsizes did nothing.
fix tight merges so they work over all colls.
fix merge counting to be fast and not loop over
all rdbbases which could be thousands.
add num mirrors to rebalance.txt.
fix updateCrawlInfo to wait for all replies. critical error!
now we store a "sharded by termid" bit in posdb
key for checksums, etc keys that are not sharded
by docid. save having to do disk seeks on every
host in the cluster to do a dup check, etc.
to allowing syncing with newly added or deleted
collections even if a host was dead when collection
was added/deleted. also added parm change request
queueing.
trying to dump to coll directories that
had been moved to trash folder.
and printing out "deleted from under us".
at least it was corrupting data in RdbMem
this time because i added m_dumpErrno logic.
disable writing was engaged. we already
had it check to see if tree was saving,
but not if writes were disabled. so it
gets ETRYAGAIN and retries later.
fix maxspidersperip limitation not being obeyed.
removed fakedb.
only add "0" time waiting tree keys to waiting tree.
only scanSpiderdb() will change their times to
a future time or add them to doledb directly.
confirmLockAcquisition() will not add to waitingtree
if max spiders per ip limit would be exceeded.
an incoming spider reply will trigger the add to
waiting tree with a time of "0".
collections, really limited by the sizeof(collnum_t) only now,
which is 16bits, 15bits unsigned, which is the limitation.
can always expand this so we can have more than 32k collections.