By extending loadRCE to accept how much the caller wants returned
rce.buf to be xincref'ed we can teach it to do the incref consistently
under rce.parent lock either itself, or schedule the incref to loadRCE
to be done after rce is loaded right before it is exposed to outside
world (rce waiters + gc).

which is showing itself as e.g.

---- 8< ----
panic: Buf: refcnt < 0

goroutine 7 [running]:
lab.nexedi.com/kirr/neo/go/zodb.(*Buf).Release(0xc42000db40)
        /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/buffer.go:107 +0x116
lab.nexedi.com/kirr/neo/go/zodb.(*Buf).XRelease(0xc42000db40)
        /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/buffer.go:136 +0x3a
lab.nexedi.com/kirr/neo/go/zodb.(*Cache).gc(0xc4200101c0)
        /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/cache.go:488 +0x22f
lab.nexedi.com/kirr/neo/go/zodb.(*Cache).gcmain(0xc4200101c0)
        /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/cache.go:451 +0x3f
created by lab.nexedi.com/kirr/neo/go/zodb.NewCache
        /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/cache.go:123 +0x126
---- 8< ----

during zhash.go -prefetch run.

This patch adds the test for this bug and explains the problem in detail
there. The test currently fails:

--- FAIL: TestCache (0.01s)
panic: Buf.Incref: refcnt was < 1 [recovered]
        panic: Buf.Incref: refcnt was < 1

goroutine 18 [running]:
testing.tRunner.func1(0xc4240ce000)
        /home/kirr/src/tools/go/go/src/testing/testing.go:711 +0x2d2
panic(0x588a20, 0x5ebe70)
        /home/kirr/src/tools/go/go/src/runtime/panic.go:491 +0x283
lab.nexedi.com/kirr/neo/go/zodb.(*Buf).Incref(0xc4200c2460)
        /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/buffer.go:134 +0x58
lab.nexedi.com/kirr/neo/go/zodb.(*Buf).XIncref(0xc4200c2460)
        /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/buffer.go:148 +0x3a
lab.nexedi.com/kirr/neo/go/zodb.(*Cache).Load(0xc4200ca000, 0x6ab1a0, 0xc42001a148, 0x4, 0x1, 0x10, 0xc4200bf2e0, 0x10, 0x10)
        /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/cache.go:168 +0x1c7
lab.nexedi.com/kirr/neo/go/zodb.TestCache.func2(0x4, 0x1, 0xc4200c2440, 0x4, 0x0, 0x0)
        /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/cache_test.go:155 +0xcc
lab.nexedi.com/kirr/neo/go/zodb.TestCache(0xc4240ce000)
        /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/cache_test.go:579 +0x4eb5
testing.tRunner(0xc4240ce000, 0x5d4728)
        /home/kirr/src/tools/go/go/src/testing/testing.go:746 +0xd0
created by testing.(*T).Run
        /home/kirr/src/tools/go/go/src/testing/testing.go:789 +0x2de
exit status 2

- IStorageDriver implements only raw access to a particular storage
- Prefetching logic + other common bits are implemented byt zodb package
  infrastructure (e.g. Prefetch() and URL()) and access to them is
  provided by IStorage.

It currently regresses plain load speed:

name                           old time/object  new time/object  delta
dataset:wczblk1-8
deco/fs1/zhash.py                  15.4µs ± 4%      15.2µs ± 2%      ~     (p=0.159 n=4+5)
deco/fs1/zhash.py-P16               116µs ±17%       121µs ±21%      ~     (p=0.407 n=16+16)
deco/fs1/zhash.go                  1.50µs ± 0%      5.30µs ± 0%  +253.33%  (p=0.016 n=5+4)
deco/fs1/zhash.go+prefetch128      4.08µs ± 5%      4.20µs ± 2%      ~     (p=0.143 n=5+5)
deco/fs1/zhash.go-P16             4.77µs ±179%     33.85µs ±54%  +610.14%  (p=0.000 n=15+16)
dataset:prod1-1024
deco/fs1/zhash.py                  12.2µs ± 1%      12.4µs ± 4%      ~     (p=0.540 n=4+5)
deco/fs1/zhash.py-P16               102µs ±12%       101µs ±16%      ~     (p=0.802 n=16+16)
deco/fs1/zhash.go                  1.10µs ± 0%      2.66µs ± 2%  +141.82%  (p=0.008 n=5+5)
deco/fs1/zhash.go+prefetch128      2.70µs ± 0%      2.67µs ± 3%      ~     (p=1.000 n=4+4)
deco/fs1/zhash.go-P16             3.20µs ±138%     17.25µs ±42%  +439.06%  (p=0.000 n=16+16)

will try to investigate what is going on inside.

also:

- DataTid -> DataTidHint + 0 if there is no such hint.

So now we are benchmarking disk for sizes 4K and 2M which are the usual
sizes popping up with e.g. wendelin.core.

* origin/master:
  client: account for cache hit/miss statistics
  client: remove redundant information from cache's __repr__
  cache: fix possible endless loop in __repr__/_iterQueue
  storage: speed up replication by not getting object next_serial for nothing
  storage: speed up replication by sending bigger network packets
  neoctl: remove ignored option
  client: bug found, add log to collect more information
  client: new 'cache-size' Storage option
  doc: mention HTTPS URLs when possible
  doc: update comment in neolog about Python issue 13773
  neolog: add support for xz-compressed logs, using external xzcat commands
  neolog: --from option now also tries to parse with dateutil
  importer: do not crash if a backup cluster tries to replicate
  storage: disable data deduplication by default
  Release version 1.8.1

This information is handy to see how well cache performs.

Amended by Julien Muchembled:
- do not abbreviate some existing field names in repr result (asking the
  user to look at the source code in order to decipher logs is not nice)
- hit: change from %.1f to %.3g
- hit: hide it completely if nload is 0
- use __future__.division instead of adding more casts to float

On my disk it gives:

name                                 time/op
deco/disk/randread/direct/4K-min     98.0µs ± 1%
deco/disk/randread/direct/4K-avg      104µs ± 0%
deco/disk/randread/direct/1M-min     2.90ms ±17%
deco/disk/randread/direct/1M-avg     3.55ms ± 0%
deco/disk/randread/pagecache/4K-min   227ns ± 1%
deco/disk/randread/pagecache/4K-avg   629ns ± 0%
deco/disk/randread/pagecache/1M-min  70.8µs ± 7%
deco/disk/randread/pagecache/1M-avg  99.4µs ± 1%

INFO Z2 Log files reopened successfully
INFO SignalHandler Caught signal SIGTERM
INFO Z2 Shutting down fast
INFO ZServer closing HTTP to new connections
ERROR ZODB.Connection Couldn't load state for BTrees.LOBTree.LOBucket 0xc12e29
Traceback (most recent call last):
  File "ZODB/Connection.py", line 909, in setstate
    self._setstate(obj, oid)
  File "ZODB/Connection.py", line 953, in _setstate
    p, serial = self._storage.load(oid, '')
  File "neo/client/Storage.py", line 81, in load
    return self.app.load(oid)[:2]
  File "neo/client/app.py", line 355, in load
    data, tid, next_tid, _ = self._loadFromStorage(oid, tid, before_tid)
  File "neo/client/app.py", line 387, in _loadFromStorage
    askStorage)
  File "neo/client/app.py", line 297, in _askStorageForRead
    self.sync()
  File "neo/client/app.py", line 898, in sync
    self._askPrimary(Packets.Ping())
  File "neo/client/app.py", line 163, in _askPrimary
    return self._ask(self._getMasterConnection(), packet,
  File "neo/client/app.py", line 177, in _getMasterConnection
    result = self.master_conn = self._connectToPrimaryNode()
  File "neo/client/app.py", line 202, in _connectToPrimaryNode
    index = (index + 1) % len(master_list)
ZeroDivisionError: integer division or modulo by zero

It's not possible yet to replicate a node that is importing data.
One must wait that the migration is finished.