This protocol version corresponds to protocol version used by NEO/py
v1.12 and was set in NEO/py commit c6453626 (Bump protocol version).

The protocol definition was updated to match that NEO/py release in
the previous patches.

This corresponds to NEO/py commit 2a27239d (tweak: add option to
simulate).

This corresponds to NEO/py commit c2c9e99d (Better error reporting from
the master to neoctl for denied requests).

This corresponds to NEO/py commit 21190ee7 (Make 'neoctl print pt'
report the number of replicas).

This corresponds to NEO/py commit ef5fc508 (Make the number of replicas
modifiable when the cluster is running).

One important change in the protocol is that Client no longer queries
Master for partition table - instead M pushed partTab to C right after
identification (after pushing nodeTab).

See also: https://neo.nexedi.com/P-NEO-Protocol.Specification.2019?portal_skin=CI_slideshow#/9/5

This corresponds to NEO/py commit 27e3f620 (New --new-nid storage option
for fast cloning).

NEO 1.12

* tag 'v1.12': (28 commits)
  Release version 1.12
  master: reject drop/tweak ctl commands that could lead to unwanted status
  qa: extend test reproducing the migration of a big ZODB to NEO
  neoctl: better display of full partition tables
  Bump protocol version
  tweak: add option to simulate
  tweak: do not crash when trying to remove all nodes
  tweak: do not touch cells of nodes that are intended to be dropped
  Better error reporting from the master to neoctl for denied requests
  Make 'neoctl print pt' report the number of replicas
  Make the number of replicas modifiable when the cluster is running
  New --new-nid storage option for fast cloning
  qa: fix 2 tests with ZODB5
  qa: new tools/stress options to evaluate MySQL engines
  qa: provide a way to let tests start 1 mysqld per storage node
  mysql: make 'user' actually optional in the DB connection string
  mysql: specify column families for RocksDB
  qa: add testIncremental (testImporter) test
  importer: fix hidden "maximum recursion depth exceeded" at startup
  importer: fix closure of ZODB, and also do it when the import is finished
  sqlite: fix resumption of migration to NEO with Importer
  qa: fix a random failure in threaded tests
  importer: speed up startup when the import is already finished
  importer: fix replication (as source) once import is finished
  storage: fix DatabaseManager.getLastTID with max_tid
  qa: remove 2 useless unit tests
  storage: allow the master to change our node id
  Rename --uuid command-line options into --nid
  importer: fix possible data loss on writeback

This protocol version corresponds to protocol version used by NEO/py
v1.11 and was set in NEO/py commit 9a5b46dd (Bump protocol version).

The protocol definition was updated to match that NEO/py release in
the previous patch.

This corresponds to NEO/py commit 64826794 (New neoctl command to flush
the logs of all nodes in the cluster).

NEO 1.11

* tag 'v1.11': (52 commits)
  Release version 1.11
  Fix short descriptions of neoctl & neomigrate in their headers
  Update copyright year
  qa: new tool to stress-test NEO
  master: fix typo in comment
  Fix error handling when setting up a listening connector
  Fix incomplete/incorrect mapping of node ids in logs
  Fix log corruption on rotation in multi-threaded applications (e.g. client)
  sqlite: optimize storage of metadata
  neolog: do not die when a table is corrupted
  neolog: add support for zstd-compressed logs
  neolog: do not hardcode default value of -L option in help message
  fixup! New log format to show node id (and optionally cluster name) in node column
  New log format to show node id (and optionally cluster name) in node column
  fixup! client: discard late answers to lockless writes
  client: fix race condition between Storage.load() and invalidations
  client: fix race condition in refcounting dispatched answer packets
  More RTMIN+2 (log) information for clients and connections
  storage: check for conflicts when notifying that the a partition is replicated
  storage: clarify several assertions
  qa: new expectedFailure testcase method
  client: merge ConnectionPool inside Application
  client: prepare merge of ConnectionPool inside Application
  client: fix AssertionError when trying to reconnect too quickly after an error
  qa: fix attributeTracker
  storage: fix storage leak when an oid is stored several times within a transaction
  client: discard late answers to lockless writes
  qa: in threaded tests, log queued packets when "tic is looping forever"
  In logs, dump the partition table in a more compact and readable way
  storage: fix write-locking bug when a deadlock happens at the end of a replication
  client: log_flush most exceptions raised from Application to ZODB
  client: fix assertion failure in case of conflict + storage disconnection
  client: simplify connection management in transaction contexts
  client: also vote to nodes that only check serials
  qa: deindent code
  Bump protocol version
  client: fix undetected disconnections to storage nodes during commit
  Fix data corruption due to undetected conflicts after storage failures
  master: notify replicating nodes of aborted watched transactions
  New neoctl command to flush the logs of all nodes in the cluster
  storage: fix premature write-locking during rebase when replication ends
  client: fix race condition when a storage connection is closed just after identification
  storage: relax assertion
  comments, unused import
  storage: fix write-lock leak
  client: fix possible corruption in case of network failure with a storage
  qa: comment about potential freeze when a functional test ends
  storage: fix assertion failure in case of connection reset with a client node
  qa: document a rare random failure in testExport
  debug: add script to trace all accesses to the client cache
  Use argparse instead of optparse
  neolog: use argparse instead of optparse
  Add comment about dormant bug when sending a lot of data to a slow node
  client: make clearer that max_size attribute is used from outside ClientCache

This protocol version corresponds to protocol version used by NEO/py v1.10.

The protocol definition was updated to match that NEO/py release in
the previous patches.

This corresponds to NEO/py commit 97af23cc (Maximize resiliency by
taking into account the topology of storage nodes).

- Rename GetObject .Tid    -> .Before
- Rename GetObject .Serial -> .At
- Sync docstrings

This corresponds to NEO/py commit 9f0f2afe (protocol: update packet
docstrings).

This corresponds to NEO/py commit 52db5607 ("protocol: a single byte is
more than enough to encode enums").

Don't skip a code when going request1->request2 through `Request1
Answer1 Request2`. For example before this patch:

	1		RequestIdentification
	1 | answerBit	AcceptIdentification
	3		Ping
	3 | answerBit	Pong
	...

after this patch:

	1		RequestIdentification
	1 | answerBit	AcceptIdentification
	2		Ping
	2 | answerBit	Pong
	...

This corresponds to NEO/py commit a00ab78b ("protocol: small cleanup in
packet registration").

Corresponds to NEO/py commit b3dd6973 ("Optimize resumption of
replication by starting from a greater TID").

Corresponds to NEO/py commit 3efbbfe3 ("master: automatically discard
feeding cells that get out-of-date").

NEO 1.10

* tag 'v1.10': (55 commits)
  Release version 1.10
  Maximize resiliency by taking into account the topology of storage nodes
  storage: also commit updated cell TID at each replicated chunk of 'obj' records
  storage: skip useless work when unlocking transactions
  qa: flush logs at the end of each test when -L is not used
  qa: add a log in case that a mysterious bug happens again
  storage: clarify log about data deletion of discarded cells
  debug: new example to run the profiler for 1 minute
  mysql: fix replication of big oids (> 16M)
  tests/cluster: speedup waiting a bit
  protocol: update packet docstrings
  Bump protocol version
  protocol: a single byte is more than enough to encode enums
  protocol: small cleanup in packet registration
  Optimize resumption of replication by starting from a greater TID
  importer: update comment about a workaround for ZODB3
  Micro-optimization of p64/u64
  qa: add a log in testBackupNodeLost for easier debugging
  Document that the bug when checking replicas may also cause the master to crash
  storage: stop logging 'Abort TXN' for txn that have been locked
  storage: split _migrate2() for reusable _alterTable()
  qa: new testStorageUpgrade
  qa: update testStorageUpgrade data for what is not automatically upgraded
  qa: original data for the future testStorageUpgrade
  sqlite: fix indexes of upgraded db
  importer: fix NameError when recovering during tpc_finish
  fixup! importer: fetch and process the data to import in a separate process
  Serialize empty transaction extension with an empty string
  client: fix partial import from a source storage
  qa: give a title to subprocesses of functional tests
  importer: give a title to the 'import' and 'writeback' subprocesses
  importer: fetch and process the data to import in a separate process
  importer: new option to write back new transactions to the source database
  importer: log when the transaction index for FileStorage DB is built
  importer: open imported zodb in read-only whenever possible
  fixup! mysql: fix remaining places where a server disconnection was not catched
  fixup! storage: speed up replication by sending bigger network packets
  mysql: do not full-scan for duplicates of big oids if deduplication is disabled
  mysql: fix remaining places where a server disconnection was not catched
  fixup! Add support for custom compression levels
  importer: reenable compression by default
  qa: review testImporter
  qa: remove a few uses of 'chr'
  Fix a few issues with ZODB5
  importer: small code cleanup in speedupFileStorageTxnLookup patch
  importer: do not trigger speedupFileStorageTxnLookup uselessly
  Add support for custom compression levels
  setup: update MANIFEST.in
  importer: do not checksum data twice
  client: store uncompressed if compressed size is equal
  fixup! master: automatically discard feeding cells that get out-of-date
  master: automatically discard feeding cells that get out-of-date
  qa: remove useless indentation in testSafeTweak
  bench: new option to mesure ZEO perfs in matrix test
  bench: reduce number of partitions in matrix test
  storage: fix replication of creation undone

ZEO4 does not have msgpack support and does not take $ZEO_MSGPACK into
account. With ZEO4 this test was failing before:

	--- FAIL: TestHandshake (0.46s)
	    --- FAIL: TestHandshake/py/msgpack=true (0.24s)
	        zeo_test.go:241: handshake: encoding=Z  ; want M

We don't have infrastructure to check python packages versions, so
check it by verifying ZEO.asyncio presence.

In that case at0 was initialized as 0 and still considered uninitialized
by flushEventq0:

    (neo) (z-dev) (g.env) kirr@deco:~/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/storage/zeo$ go test -run Empty
    ------
    2020-10-12T07:39:25 INFO ZEO.runzeo (146240) opening storage '1' using FileStorage
    ------
    2020-10-12T07:39:25 INFO ZEO.StorageServer StorageServer created RW with storages: 1:RW:/tmp/zeo905263273/1.fs
    ------
    2020-10-12T07:39:25 INFO ZEO.asyncio.server listening on /tmp/zeo905263273/1.fs.zeosock
    ------
    2020-10-12T07:39:25 INFO ZEO.asyncio.base Connected server protocol
    ------
    2020-10-12T07:39:25 INFO ZEO.asyncio.server received handshake 'Z5'
    2020/10/12 07:39:25 /tmp/zeo905263273/1.fs.zeosock: EOF
    --- FAIL: TestEmptyDB (0.22s)
        --- FAIL: TestEmptyDB/py/msgpack=false (0.22s)
    panic: flush, but .at0 not yet initialized [recovered]
            panic: flush, but .at0 not yet initialized

    goroutine 7 [running]:
    testing.tRunner.func1.1(0x644a60, 0x6e1a50)
            /home/kirr/src/tools/go/go/src/testing/testing.go:1072 +0x30d
    testing.tRunner.func1(0xc000001e00)
            /home/kirr/src/tools/go/go/src/testing/testing.go:1075 +0x41a
    panic(0x644a60, 0x6e1a50)
            /home/kirr/src/tools/go/go/src/runtime/panic.go:969 +0x175
    lab.nexedi.com/kirr/neo/go/zodb/storage/zeo.(*zeo).flushEventq0(0xc00018a000)
            /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/storage/zeo/zeo.go:180 +0xf3
    lab.nexedi.com/kirr/neo/go/zodb/storage/zeo.openByURL(0x6e9ca0, 0xc000016108, 0xc000138120, 0xc000153d98, 0x0, 0x0, 0x0, 0x0, 0x0)
            /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/storage/zeo/zeo.go:488 +0x5ba
    lab.nexedi.com/kirr/neo/go/zodb/storage/zeo.zeoOpen(0xc000018740, 0x1e, 0xc000049d98, 0x0, 0x0, 0x0, 0x0)
            /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/storage/zeo/zeo_test.go:285 +0x17b
    lab.nexedi.com/kirr/neo/go/zodb/storage/zeo.withZEO.func1(0xc000001e00, 0x6e9ea0, 0xc00005e6c0)
            /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/storage/zeo/zeo_test.go:219 +0xd0
    lab.nexedi.com/kirr/neo/go/zodb/storage/zeo.withZEOSrv.func2.1(0xc0000185c0, 0x16)
            /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/storage/zeo/zeo_test.go:205 +0xfb
    lab.nexedi.com/kirr/neo/go/zodb/storage/zeo.withZEOSrv.func1(0xc000001e00, 0xc00000e5a0)
            /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/storage/zeo/zeo_test.go:185 +0x129
    lab.nexedi.com/kirr/neo/go/zodb/storage/zeo.withZEOSrv.func2(0xc000001e00)
            /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/storage/zeo/zeo_test.go:197 +0x105
    testing.tRunner(0xc000001e00, 0xc00000e440)
            /home/kirr/src/tools/go/go/src/testing/testing.go:1123 +0xef
    created by testing.(*T).Run
            /home/kirr/src/tools/go/go/src/testing/testing.go:1168 +0x2b3
    exit status 2
    FAIL    lab.nexedi.com/kirr/neo/go/zodb/storage/zeo     0.227s

-> Fix it by using dedicated field marking whether .at0 was initialized or not yet.

For virtio NICs /sys/class/net/<NIC>/device lead to $pcidev/virtioX, not just $pcidev, e.g.:

	$ realpath /sys/class/net/ens3/device
	/sys/devices/pci0000:00/0000:00:03.0/virtio0

and we were extracting virtio0 instead of 0000:00:03.0 as PCI device identifier.

-> Fix it by recognizing and stripping /virtioX suffix.

For example under KVM it was failing as

    cpu:	Intel(R) Xeon(R) CPU E5-2678 v3 @ 2.50GHz
    cat: /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq: No such file or directory
      File "<string>", line 1
        print '%.2fGHz' % ( / 1E6)
                            ^
    SyntaxError: invalid syntax

    $ lscpu
    Architecture:        x86_64
    CPU op-mode(s):      32-bit, 64-bit
    Byte Order:          Little Endian
    Address sizes:       40 bits physical, 48 bits virtual
    CPU(s):              40
    On-line CPU(s) list: 0-39
    Thread(s) per core:  1
    Core(s) per socket:  1
    Socket(s):           40
    NUMA node(s):        1
    Vendor ID:           GenuineIntel
    CPU family:          6
    Model:               63
    Model name:          Intel(R) Xeon(R) CPU E5-2678 v3 @ 2.50GHz
    Stepping:            2
    CPU MHz:             2494.238
    BogoMIPS:            4988.47
    Virtualization:      VT-x
    Hypervisor vendor:   KVM
    Virtualization type: full
    L1d cache:           32K
    L1i cache:           32K
    L2 cache:            4096K
    L3 cache:            16384K
    NUMA node0 CPU(s):   0-39
    Flags:               fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon rep_good nopl xtopology cpuid tsc_known_freq pni pclmulqdq vmx ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm invpcid_single pti tpr_shadow vnmi flexpriority ept vpid fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid xsaveopt arat

-> Fix it by checking whether cpufreq/cpuidle directories are available, and display "?" if they are not.

Starting from Go1.14 defer is no longer slow:
https://golang.org/doc/go1.14#runtime

Starting from Go1.14 defer is no longer slow:
https://golang.org/doc/go1.14#runtime

This turns zeo tests to pass.
The bug is reported upstream here: https://github.com/zopefoundation/ZODB/issues/318

If xres fails to convert to tuple, we should error about xres, not res
which is always tuple(nil).

Receive invalidation from server and send corresponding events to watchq.
Care to send only events with tid > at0 that we initially returned when opening.

Tests pass, but they need https://github.com/zopefoundation/ZEO/pull/160

This will be used in the next patch to handle invalidateTransaction
messages.

ZEO5 adds way for messages to be encoded via either pickles or MessagePack.
However until now we were always using pickles.
Let's add msgpack support to be able to e.g. use wire encoding that
server prefers.

MsgPack support is almost fully localized in encoding.
We use tinylib/msgp runtime routines to decode/encode msg fields with known types,
and shamaton/msgpack to decode/encode msg.arg, which is interface{},
because msgp does not generally work for arbitrary reflections.

For msgpack=true, tests state is the same as with pickles: handshake
works, but load fails when verifying that Load returns correct error for
deleted object:

    TestLoad/py/msgpack=false: xtesting.go:272: load 0285cbacc06d3a4c:0000000000000007: returned err unexpected:
        have: /tmp/zeo170183943/1.fs.zeosock: load 0285cbacc06d3a4c:0000000000000007: 0000000000000007: no such object
        want: /tmp/zeo170183943/1.fs.zeosock: load 0285cbacc06d3a4c:0000000000000007: 0000000000000007: object was deleted @0285cbacc06d3a4c
    TestLoad/py/msgpack=false: xtesting.go:272: load 0285cbad858bf2e6:0000000000000006: returned err unexpected:
        have: /tmp/zeo170183943/1.fs.zeosock: load 0285cbad858bf2e6:0000000000000006: 0000000000000006: no such object
        want: /tmp/zeo170183943/1.fs.zeosock: load 0285cbad858bf2e6:0000000000000006: 0000000000000006: object was deleted @0285cbad858bf2e6
    TestLoad/py/msgpack=false: xtesting.go:290: load 7fffffffffffffff:0000000000000007: returned err unexpected:
        have: /tmp/zeo170183943/1.fs.zeosock: load 7fffffffffffffff:0000000000000007: 0000000000000007: no such object
        want: /tmp/zeo170183943/1.fs.zeosock: load 7fffffffffffffff:0000000000000007: 0000000000000007: object was deleted @0285cbacc06d3a4c
    TestLoad/py/msgpack=false: xtesting.go:290: load 7fffffffffffffff:0000000000000006: returned err unexpected:
        have: /tmp/zeo170183943/1.fs.zeosock: load 7fffffffffffffff:0000000000000006: 0000000000000006: no such object
        want: /tmp/zeo170183943/1.fs.zeosock: load 7fffffffffffffff:0000000000000006: 0000000000000006: object was deleted @0285cbad858bf2e6

    TestLoad/py/msgpack=true: xtesting.go:272: load 0285cbacc06d3a4c:0000000000000007: returned err unexpected:
        have: /tmp/zeo247652538/1.fs.zeosock: load 0285cbacc06d3a4c:0000000000000007: 0000000000000007: no such object
        want: /tmp/zeo247652538/1.fs.zeosock: load 0285cbacc06d3a4c:0000000000000007: 0000000000000007: object was deleted @0285cbacc06d3a4c
    TestLoad/py/msgpack=true: xtesting.go:272: load 0285cbad858bf2e6:0000000000000006: returned err unexpected:
        have: /tmp/zeo247652538/1.fs.zeosock: load 0285cbad858bf2e6:0000000000000006: 0000000000000006: no such object
        want: /tmp/zeo247652538/1.fs.zeosock: load 0285cbad858bf2e6:0000000000000006: 0000000000000006: object was deleted @0285cbad858bf2e6
    TestLoad/py/msgpack=true: xtesting.go:290: load 7fffffffffffffff:0000000000000007: returned err unexpected:
        have: /tmp/zeo247652538/1.fs.zeosock: load 7fffffffffffffff:0000000000000007: 0000000000000007: no such object
        want: /tmp/zeo247652538/1.fs.zeosock: load 7fffffffffffffff:0000000000000007: 0000000000000007: object was deleted @0285cbacc06d3a4c
    TestLoad/py/msgpack=true: xtesting.go:290: load 7fffffffffffffff:0000000000000006: returned err unexpected:
        have: /tmp/zeo247652538/1.fs.zeosock: load 7fffffffffffffff:0000000000000006: 0000000000000006: no such object
        want: /tmp/zeo247652538/1.fs.zeosock: load 7fffffffffffffff:0000000000000006: 0000000000000006: object was deleted @0285cbad858bf2e6

This is due to https://github.com/zopefoundation/ZODB/issues/318

It is already documented in pktDecodeZ comment that flags is int|bool.
However until now we were decoding it only as int.

As is it was working, but it will fail when receiving e.g.
invalidateTransaction message, because ZEO/py server actually uses bool
when sending it:

    https://github.com/zopefoundation/ZEO/blob/5.2.1-20-gcb26281d/src/ZEO/asyncio/base.py#L139

Fix it.
This will be covered by watch tests, when watch support will be added in a followup patch.

Handle pickled lists as valid input when decoding tuples.

This is more uniform and will be followed by all data types in the next patch.
Here rename tid/oid pack/unpack routines correspondingly.
Add docstrings for X=tid|oid.

Keep information about which message encoding is used on the wire in
encoding type. Make pktDecode/pktEncode and data type conversion
utilities be methods of this type. For now there is only 'pickles'
encoding, but we'll soon introduce 'msgpack'.

Currently not everything related to pickles is localized in encoding -
we'll be moving more bits to encoding in the followup patches.

First step:

- move msg and msgFlags
- move pktDecode
- move functions to pack/unpack tid and oid

more to come.