This patch logically continues previous change `bigfile/virtmem:
Introduce "mmap overlay" mode` and exposes mmap-overlay functionality to
Python: if PyBigFile backend provides .blkmmapper PyCapsule the
mmap-related methods will be extracted from it and passed on through to
virtmem - see _bigfile.h for details.

ZBigFile will use this to hook into using WCFS.

with the intention to later use WCFS through it.

Before this patch virtmem had only one mode: a BigFile backend was
providing loadblk and storeblk methods, and on every block access
loadblk was called to load block data into allocated RAM page.

However with WCFS virtmem won't be needed to do anything to load data -
because loading from head/bigfile/f mmaped through OS will be handled by
OS directly. Thus for wcfs, that leaves virtmem only to handle dirtying
and writeout.

-> Introduce "mmap overlay" mode into virtmem to handle WCFS-like
BigFile backends - that can provide read-only base layer suitable for
mmapping.

This patch is organized as follows:

- fileh_open is added flags argument to indicate which mode to use for
  opened fileh. BigFileH is added .mmap_overlay bitfield correspondingly.
  (virtmem.h)

- struct bigfile_ops is extended with 3 optional methods that a BigFile
  backend might provide to support mmap-overlay mode:

  * mmap_setup_read,
  * remmap_blk_read, and
  * munmap

  (see file.h changes for documentation of this new interface)

- if opened with MMAP_OVERLAY flag, virtmem is using those methods to
  organize VMA views backed by read-only base mmap layer and writeout
  for such VMAs (virtmem.c)

- a test is added to exercise MMAP_OVERLAY virtmem mode (test_virtmem.c)

- everything else, including bigfile.py, is switched to use
  DONT_MMAP_OVERLAY unconditionally for now.

In internal comments inside virtmem new mode is interchangeable called
"mmap overlay" and "wcfs", even though wcfs is not hooked to be used
mmap-overlaying yet.

Some preliminary history:

kirr/wendelin.core@fb6932a2    X Split PAGE_LOADED -> PAGE_LOADED, PAGE_LOADED_FOR_WRITE
kirr/wendelin.core@4a20a573    X Settled on what should happen after writeout for wcfs case
kirr/wendelin.core@f084ff9b    X Transition to all VMA under 1 fileh to be either all based on wcfs or all based on !wcfs

This patch follows-up on previous patch, that added server-side part of
isolation protocol handling, and adds client package that takes care about
WCFS isolation protocol details and provides to clients simple interface to
isolated view of bigfile data on WCFS similar to regular files: given a
particular revision of database @at, it provides synthetic read-only bigfile
memory mappings with data corresponding to @at state, but using /head/bigfile/*
most of the time to build and maintain the mappings.

The patch is organized as follows:

- wcfs.h and wcfs.cpp brings in usage documentation, internal overview and the
  main part of the implementation.

- wcfs/client/client_test.py is tests.

- The rest of the changes in wcfs/client/ are to support the implementation and tests.

Quoting package documentation for the reference:

---- 8< ----

Package wcfs provides WCFS client.

This client package takes care about WCFS isolation protocol details and
provides to clients simple interface to isolated view of bigfile data on
WCFS similar to regular files: given a particular revision of database @at,
it provides synthetic read-only bigfile memory mappings with data
corresponding to @at state, but using /head/bigfile/* most of the time to
build and maintain the mappings.

For its data a mapping to bigfile X mostly reuses kernel cache for
/head/bigfile/X with amount of data not associated with kernel cache for
/head/bigfile/X being proportional to δ(bigfile/X, at..head). In the usual
case where many client workers simultaneously serve requests, their database
views are a bit outdated, but close to head, which means that in practice
the kernel cache for /head/bigfile/* is being used almost 100% of the time.

A mapping for bigfile X@at is built from OS-level memory mappings of
on-WCFS files as follows:

                                          ___        /@revA/bigfile/X
        __                                           /@revB/bigfile/X
               _                                     /@revC/bigfile/X
                           +                         ...
     ───  ───── ──────────────────────────   ─────   /head/bigfile/X

where @revR mmaps are being dynamically added/removed by this client package
to maintain X@at data view according to WCFS isolation protocol(*).

API overview

 - `WCFS` represents filesystem-level connection to wcfs server.
 - `Conn` represents logical connection that provides view of data on wcfs
   filesystem as of particular database state.
 - `FileH` represent isolated file view under Conn.
 - `Mapping` represents one memory mapping of FileH.

A path from WCFS to Mapping is as follows:

 WCFS.connect(at)                    -> Conn
 Conn.open(foid)                     -> FileH
 FileH.mmap([blk_start +blk_len))    -> Mapping

A connection can be resynced to another database view via Conn.resync(at').

Documentation for classes provides more thorough overview and API details.

--------

(*) see wcfs.go documentation for WCFS isolation protocol overview and details.

.

Wcfs client organization
~~~~~~~~~~~~~~~~~~~~~~~~

Wcfs client provides to its users isolated bigfile views backed by data on
WCFS filesystem. In the absence of Isolation property, wcfs client would
reduce to just directly using OS-level file wcfs/head/f for a bigfile f. On
the other hand there is a simple, but inefficient, way to support isolation:
for @at database view of bigfile f - directly use OS-level file wcfs/@at/f.
The latter works, but is very inefficient because OS-cache for f data is not
shared in between two connections with @at1 and @at2 views. The cache is
also lost when connection view of the database is resynced on transaction
boundary. To support isolation efficiently, wcfs client uses wcfs/head/f
most of the time, but injects wcfs/@revX/f parts into mappings to maintain
f@at view driven by pin messages that wcfs server sends to client in
accordance to WCFS isolation protocol(*).

Wcfs server sends pin messages synchronously triggered by access to mmaped
memory. That means that a client thread, that is accessing wcfs/head/f mmap,
is completely blocked while wcfs server sends pins and waits to receive acks
from all clients. In other words on-client handling of pins has to be done
in separate thread, because wcfs server can also send pins to client that
triggered the access.

Wcfs client implements pins handling in so-called "pinner" thread(+). The
pinner thread receives pin requests from wcfs server via watchlink handle
opened through wcfs/head/watch. For every pin request the pinner finds
corresponding Mappings and injects wcfs/@revX/f parts via Mapping._remmapblk
appropriately.

The same watchlink handle is used to send client-originated requests to wcfs
server. The requests are sent to tell wcfs that client wants to observe a
particular bigfile as of particular revision, or to stop watching it.
Such requests originate from regular client threads - not pinner - via entry
points like Conn.open, Conn.resync and FileH.close.

Every FileH maintains fileh._pinned {} with currently pinned blk -> rev. This
dict is updated by pinner driven by pin messages, and is used when
new fileh Mapping is created (FileH.mmap).

In wendelin.core a bigfile has semantic that it is infinite in size and
reads as all zeros beyond region initialized with data. Memory-mapping of
OS-level files can also go beyond file size, however accessing memory
corresponding to file region after file.size triggers SIGBUS. To preserve
wendelin.core semantic wcfs client mmaps-in zeros for Mapping regions after
wcfs/head/f.size. For simplicity it is assumed that bigfiles only grow and
never shrink. It is indeed currently so, but will have to be revisited
if/when wendelin.core adds bigfile truncation. Wcfs client restats
wcfs/head/f at every transaction boundary (Conn.resync) and remembers f.size
in FileH._headfsize for use during one transaction(%).

--------

(*) see wcfs.go documentation for WCFS isolation protocol overview and details.
(+) currently, for simplicity, there is one pinner thread for each connection.
    In the future, for efficiency, it might be reworked to be one pinner thread
    that serves all connections simultaneously.
(%) see _headWait comments on how this has to be reworked.

Wcfs client locking organization

Wcfs client needs to synchronize regular user threads vs each other and vs
pinner. A major lock Conn.atMu protects updates to changes to Conn's view of
the database. Whenever atMu.W is taken - Conn.at is changing (Conn.resync),
and contrary whenever atMu.R is taken - Conn.at is stable (roughly speaking
Conn.resync is not running).

Similarly to wcfs.go(*) several locks that protect internal data structures
are minor to Conn.atMu - they need to be taken only under atMu.R (to
synchronize e.g. multiple fileh open running simultaneously), but do not
need to be taken at all if atMu.W is taken. In data structures such locks
are noted as follows

     sync::Mutex xMu;    // atMu.W  |  atMu.R + xMu

After atMu, Conn.filehMu protects registry of opened file handles
(Conn._filehTab), and FileH.mmapMu protects registry of created Mappings
(FileH.mmaps) and FileH.pinned.

Several locks are RWMutex instead of just Mutex not only to allow more
concurrency, but, in the first place for correctness: pinner thread being
core element in handling WCFS isolation protocol, is effectively invoked
synchronously from other threads via messages coming through wcfs server.
For example Conn.resync sends watch request to wcfs server and waits for the
answer. Wcfs server, in turn, might send corresponding pin messages to the
pinner and _wait_ for the answer before answering to resync:

       - - - - - -
      |       .···|·····.        ---->   = request
         pinner <------.↓        <····   = response
      |           |   wcfs
         resync -------^↓
      |      `····|·····
       - - - - - -
      client process

This creates the necessity to use RWMutex for locks that pinner and other
parts of the code could be using at the same time in synchronous scenarios
similar to the above. This locks are:

     - Conn.atMu
     - Conn.filehMu

Note that FileH.mmapMu is regular - not RW - mutex, since nothing in wcfs
client calls into wcfs server via watchlink with mmapMu held.

The ordering of locks is:

     Conn.atMu > Conn.filehMu > FileH.mmapMu

The pinner takes the following locks:

     - wconn.atMu.R
     - wconn.filehMu.R
     - fileh.mmapMu (to read .mmaps  +  write .pinned)

(*) see "Wcfs locking organization" in wcfs.go

Handling of fork

When a process calls fork, OS copies its memory and creates child process
with only 1 thread. That child inherits file descriptors and memory mappings
from parent. To correctly continue using Conn, FileH and Mappings, the child
must recreate pinner thread and reconnect to wcfs via reopened watchlink.
The reason here is that without reconnection - by using watchlink file
descriptor inherited from parent - the child would interfere into
parent-wcfs exchange and neither parent nor child could continue normal
protocol communication with WCFS.

For simplicity, since fork is seldomly used for things besides followup
exec, wcfs client currently takes straightforward approach by disabling
mappings and detaching from WCFS server in the child right after fork. This
ensures that there is no interference into parent-wcfs exchange should child
decide not to exec and to continue running in the forked thread. Without
this protection the interference might come even automatically via e.g.
Python GC -> PyFileH.__del__ -> FileH.close -> message to WCFS.

----------------------------------------

Some preliminary history:

kirr/wendelin.core@a8fa9178    X wcfs: move client tests into client/
kirr/wendelin.core@990afac1    X wcfs/client: Package overview (draft)
kirr/wendelin.core@3f83469c    X wcfs: client: Handle fork
kirr/wendelin.core@0ed6b8b6    fixup! X wcfs: client: Handle fork
kirr/wendelin.core@24378c46    X wcfs: client: Provide Conn.at()

Via custom isolation protocol that both server and clients must cooperatively
follow. This is the core change that enables file cache to be practically
shared while each client can still be provided with isolated view of the database.

This patch brings only server changes, tests + the minimum client bits to support the tests.
The client library, that will implement isolation protocol on client side, will come next.

This patch is organized as follows:

- wcfs.go brings in description of the protocol, overview of how server
  implements that protocol and the implementation itself.
  See also notes.txt

- wcfs_test.py brings in tests for server implementation.
  tWCFS._abort_ontimeout had to be moved into nogil mode into wcfs_test.pyx
  to avoid deadlock on the GIL (see comments in wcfs_test.pyx for details).

- files added in wcfs/client/ are needed to provide client-side
  implementation of WatchLink - the message exchange protocol over
  opened head/watch file - for tests. Client-side watchlink implementation
  lives in wcfs/client/wcfs_watchlink.{h,cpp}. The other additions in
  wcfs/client/ are to support that and to expose the WatchLink to Python.

  Client-side bits are done right in C++ because upcoming WCFS client
  library will be implemented in C++ to work in nogil mode in order to
  avoid deadlock on the GIL because client-side pinner thread might be
  woken-up synchronously by WCFS server at any moment, including when
  another client thread already holds the GIL and is paused by WCFS.

Some preliminary history:

kirr/wendelin.core@9b4a42a3    X invalidation design draftly settled
kirr/wendelin.core@27d91d47    X δFtail settled
kirr/wendelin.core@c27c1940    X mmap over under pagefault to this mmapping works
kirr/wendelin.core@d36b171f    X ptrace when client is under pagefault or syscall won't work
kirr/wendelin.core@c1f5bb19    X notes on why lazy-invalidate approach was taken
kirr/wendelin.core@4fbdd270    X Proof that that it is possible to change mmapping while under pagefault to it
kirr/wendelin.core@33e0dfce    X ΔTail draftly done
kirr/wendelin.core@12628943    X make sure "bye" is always processed immediately - even if a handleWatch is currently blocked
kirr/wendelin.core@af0a64cb    X test for "bye" canceling blocked handlers
kirr/wendelin.core@996dc6a8    X Fix race in test
kirr/wendelin.core@43915fe9    X wcfs: Don't forbid simultaneous watch requests
kirr/wendelin.core@941dc54b    X wcfs: threading.Lock -> sync.Mutex
kirr/wendelin.core@d75b2304    X wcfs: Move _abort_ontimeout to pyx/nogil
kirr/wendelin.core@79234659    X Notes on why eagier invalidation was rejected
kirr/wendelin.core@f05271b1    X Test that sysread(/head/watch) can be interrupted
kirr/wendelin.core@5ba816da    X restore test_wcfs_watch_robust after f05271b1.
kirr/wendelin.core@4bd88564    X "Invalidation protocol" -> "Isolation protocol"
kirr/wendelin.core@f7b54ca4    X avoid fmt::vsprintf  (now compils again with latest pygolang@master)
kirr/wendelin.core@0a8fcd9d    X wcfs/client: Move EOF -> pygolang
kirr/wendelin.core@153e02e6    X test_wcfs_watch_setup and test_wcfs_watch_setup_ahead work again
kirr/wendelin.core@17f98edc    X wcfs: client: os: Factor syserr -> string into _sysErrString
kirr/wendelin.core@7b0c301c    X wcfs: tests: Fix tFile.assertBlk not to segfault on a test failure
kirr/wendelin.core@b74dda09    X Start switching Track from Track(key) to Track(keycov)
kirr/wendelin.core@8b5d8523    X Move tracking of which blocks were accessed from wcfs to ΔFtail

Use ΔFtail.Track on every READ, and query accumulated ΔFtail upon
receiving ZODB invalidation to query it about which blocks of which
files have been changed. Then invalidate those blocks in OS file cache.

See added documentation to wcfs.go and notes.txt for details.

Now the filesystem is no longer stale: it provides view of data
that is uptodate wrt changes on ZODB storage.

Some preliminary history:

kirr/wendelin.core@9b4a42a3    X invalidation design draftly settled
kirr/wendelin.core@27d91d47    X δFtail settled
kirr/wendelin.core@33e0dfce    X ΔTail draftly done
kirr/wendelin.core@822366a7    X keeping fd to root opened prevents the filesystem from being unmounted
kirr/wendelin.core@89ad3a79    X Don't keep ZBigFile activated during whole current transaction
kirr/wendelin.core@245511ac    X Give pointer on from where to get nxd-fuse.ko
kirr/wendelin.core@d1cd128c    X Hit FUSE-related deadlock
kirr/wendelin.core@d134ee44    X FUSE lookup deadlock should be hopefully fixed
kirr/wendelin.core@0e60e9ff    X wcfs: Don't noise ZWatcher trace logs with "select ..."
kirr/wendelin.core@bf9a7405    X No longer rely on ZODB cache invariant for invalidations

FileSock is bidirectional channel associated with opened file.

FileSock provides streaming write/read operations for filesystem server that
are correspondingly matched with read/write operations on filesystem user side.

WCFS will use FileSock to implement exchange over .wcfs/zhead and,
later, head/watch files.

Some preliminary history:

kirr/wendelin.core@b17aeb8c    X Change FileSock to use xio.Pipe which is io.Pipe + support for IO cancellation

ΔFtail builds on ΔBtail and  provides ZBigFile-level history that WCFS
will use to compute which blocks of a ZBigFile need to be invalidated in
OS file cache given raw ZODB changes on ZODB invalidation message.

It also will be used by WCFS to implement isolation protocol, where on
every FUSE READ request WCFS will query ΔFtail to find out revision of
corresponding file block.

Quoting ΔFtail documentation:

---- 8< ----

ΔFtail provides ZBigFile-level history tail.

It translates ZODB object-level changes to information about which blocks of
which ZBigFile were modified, and provides service to query that information.

ΔFtail class documentation
~~~~~~~~~~~~~~~~~~~~~~~~~~

ΔFtail represents tail of revisional changes to files.

It semantically consists of

    []δF			; rev ∈ (tail, head]

where δF represents a change in files space

    δF:
    	.rev↑
    	{} file ->  {}blk | EPOCH

Only files and blocks explicitly requested to be tracked are guaranteed to
be present. In particular a block that was not explicitly requested to be
tracked, even if it was changed in δZ, is not guaranteed to be present in δF.

After file epoch (file creation, deletion, or any other change to file
object) previous track requests for that file become forgotten and have no
further effect.

ΔFtail provides the following operations:

  .Track(file, blk, path, zblk)	- add file and block reached via BTree path to tracked set.

  .Update(δZ) -> δF				- update files δ tail given raw ZODB changes
  .ForgetPast(revCut)			- forget changes ≤ revCut
  .SliceByRev(lo, hi) -> []δF		- query for all files changes with rev ∈ (lo, hi]
  .SliceByFileRev(file, lo, hi) -> []δfile	- query for changes of a file with rev ∈ (lo, hi]
  .BlkRevAt(file, #blk, at) -> blkrev	- query for what is last revision that changed
    					  file[#blk] as of @at database state.

where δfile represents a change to one file

    δfile:
    	.rev↑
    	{}blk | EPOCH

See also zodb.ΔTail and xbtree.ΔBtail

Concurrency

ΔFtail is safe to use in single-writer / multiple-readers mode. That is at
any time there should be either only sole writer, or, potentially several
simultaneous readers. The table below classifies operations:

    Writers:  Update, ForgetPast
    Readers:  Track + all queries (SliceByRev, SliceByFileRev, BlkRevAt)

Note that, in particular, it is correct to run multiple Track and queries
requests simultaneously.

ΔFtail organization
~~~~~~~~~~~~~~~~~~~

ΔFtail leverages:

    - ΔBtail to track changes to ZBigFile.blktab BTree, and
    - ΔZtail to track changes to ZBlk objects and to ZBigFile object itself.

then every query merges ΔBtail and ΔZtail data on the fly to provide
ZBigFile-level result.

Merging on the fly, contrary to computing and maintaining vδF data, is done
to avoid complexity of recomputing vδF when tracking set changes. Most of
ΔFtail complexity is, thus, located in ΔBtail, which implements BTree diff
and handles complexity of recomputing vδB when set of tracked blocks
changes after new track requests.

Changes to ZBigFile object indicate epochs. Epochs could be:

    - file creation or deletion,
    - change of ZBigFile.blksize,
    - change of ZBigFile.blktab to point to another BTree.

Epochs represent major changes to file history where file is assumed to
change so dramatically, that practically it can be considered to be a
"whole" change. In particular, WCFS, upon seeing a ZBigFile epoch,
invalidates all data in corresponding OS-level cache for the file.

The only historical data, that ΔFtail maintains by itself, is history of
epochs. That history does not need to be recomputed when more blocks become
tracked and is thus easy to maintain. It also can be maintained only in
ΔFtail because ΔBtail and ΔZtail does not "know" anything about ZBigFile.

Concurrency

In order to allow multiple Track and queries requests to be served in
parallel, ΔFtail bases its concurrency promise on ΔBtail guarantees +
snapshot-style access for vδE and ztrackInBlk in queries:

1. Track calls ΔBtail.Track and quickly updates .byFile, .byRoot and
   _RootTrack indices under a lock.

2. BlkRevAt queries ΔBtail.GetAt and then combines retrieved information
   about zblk with vδE and δZ.

3. SliceByFileRev queries ΔBtail.SliceByRootRev and then merges retrieved
   vδT data with vδZ, vδE and ztrackInBlk.

4. In queries vδE is retrieved/built in snapshot style similarly to how vδT
   is built in ΔBtail. Note that vδE needs to be built only the first time,
   and does not need to be further rebuilt, so the logic in ΔFtail is simpler
   compared to ΔBtail.

5. for ztrackInBlk - that is used by SliceByFileRev query - an atomic
   snapshot is retrieved for objects of interest. This allows to hold
   δFtail.mu lock for relatively brief time without blocking other parallel
   Track/queries requests for long.

Combined this organization allows non-overlapping queries/track-requests
to run simultaneously. (This property is essential to WCFS because otherwise
WCFS would not be able to serve several non-overlapping READ requests to one
file in parallel.)

See also "Concurrency" in ΔBtail organization for more details.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Some preliminary history:

kirr/wendelin.core@ef74aebc    X ΔFtail: Keep reference to ZBigFile via Oid, not via *ZBigFile
kirr/wendelin.core@bf9a7405    X No longer rely on ZODB cache invariant for invalidations
kirr/wendelin.core@46340069    X found by Random
kirr/wendelin.core@e7b598c6    X start of ΔFtail.SliceByFileRev rework to function via merging δB and δZ histories on the fly
kirr/wendelin.core@59c83009    X ΔFtail.SliceByFileRoot tests started to work draftly after "on-the-fly" rework
kirr/wendelin.core@210e9b07    X Fix ΔBtail.SliceByRootRev (lo,hi] handling
kirr/wendelin.core@bf3ace66    X ΔFtail: Rebuild vδE after first track
kirr/wendelin.core@46624787    X ΔFtail: `go test -failfast -short -v -run Random -randseed=1626793016249041295` discovered problems
kirr/wendelin.core@786dd336    X Size no longer tracks [0,∞) since we start tracking when zfile is non-empty
kirr/wendelin.core@4f707117    X test that shows problem of SliceByRootRev where untracked blocks are not added uniformly into whole history
kirr/wendelin.core@c0b7e4c3    X ΔFtail.SliceByFileRev: Fix untracked entries to be present uniformly in result
kirr/wendelin.core@aac37c11    X zdata: Introduce T to start removing duplication in tests
kirr/wendelin.core@bf411aa9    X zdata: Deduplicate zfile loading
kirr/wendelin.core@b74dda09    X Start switching Track from Track(key) to Track(keycov)
kirr/wendelin.core@aa0288ce    X Switch SliceByRootRev to vδTSnapForTracked
kirr/wendelin.core@588a512a    X zdata: Switch SliceByFileRev not to clone Zinblk
kirr/wendelin.core@8b5d8523    X Move tracking of which blocks were accessed from wcfs to ΔFtail
kirr/wendelin.core@30f5ddc7    ΔFtail += .Epoch in δf
kirr/wendelin.core@22f5f096    X Rework ΔFtail so that BlkRevAt works with ZBigFile checkout from any at ∈ (tail, head]
kirr/wendelin.core@0853cc9f    X ΔFtail + tests
kirr/wendelin.core@124688f9    X ΔFtail fixes
kirr/wendelin.core@d85bb82c    ΔFtail concurrency

ΔBtail provides BTree-level history tail that WCFS - via ΔFtail - will
use to compute which blocks of a ZBigFile need to be invalidated in OS
file cache given raw ZODB changes on ZODB invalidation message.

It also will be used by WCFS to implement isolation protocol, where on
every FUSE READ request WCFS will query ΔBtail - again via ΔFtail - to
find out revision of corresponding file block.

Quoting ΔBtail documentation:

---- 8< ----

ΔBtail provides BTree-level history tail.

It translates ZODB object-level changes to information about which keys of
which BTree were modified, and provides service to query that information.

ΔBtail class documentation
~~~~~~~~~~~~~~~~~~~~~~~~~~

ΔBtail represents tail of revisional changes to BTrees.

It semantically consists of

    []δB			; rev ∈ (tail, head]

where δB represents a change in BTrees space

    δB:
    	.rev↑
    	{} root -> {}(key, δvalue)

It covers only changes to keys from tracked subset of BTrees parts.
In particular a key that was not explicitly requested to be tracked, even if
it was changed in δZ, is not guaranteed to be present in δB.

ΔBtail provides the following operations:

  .Track(path)	- start tracking tree nodes and keys; root=path[0], keys=path[-1].(lo,hi]

  .Update(δZ) -> δB				- update BTree δ tail given raw ZODB changes
  .ForgetPast(revCut)			- forget changes ≤ revCut
  .SliceByRev(lo, hi) -> []δB		- query for all trees changes with rev ∈ (lo, hi]
  .SliceByRootRev(root, lo, hi) -> []δT	- query for changes of a tree with rev ∈ (lo, hi]
  .GetAt(root, key, at) -> (value, rev)	- get root[key] @at assuming root[key] ∈ tracked

where δT represents a change to one tree

    δT:
    	.rev↑
    	{}(key, δvalue)

An example for tracked set is a set of visited BTree paths.
There is no requirement that tracked set belongs to only one single BTree.

See also zodb.ΔTail and zdata.ΔFtail

Concurrency

ΔBtail is safe to use in single-writer / multiple-readers mode. That is at
any time there should be either only sole writer, or, potentially several
simultaneous readers. The table below classifies operations:

    Writers:  Update, ForgetPast
    Readers:  Track + all queries (SliceByRev, SliceByRootRev, GetAt)

Note that, in particular, it is correct to run multiple Track and queries
requests simultaneously.

ΔBtail organization
~~~~~~~~~~~~~~~~~~~

ΔBtail keeps raw ZODB history in ΔZtail and uses BTree-diff algorithm(*) to
turn δZ into BTree-level diff. For each tracked BTree a separate ΔTtail is
maintained with tree-level history in ΔTtail.vδT .

Because it is very computationally expensive(+) to find out for an object to
which BTree it belongs, ΔBtail cannot provide full BTree-level history given
just ΔZtail with δZ changes. Due to this ΔBtail requires help from
users, which are expected to call ΔBtail.Track(treepath) to let ΔBtail know
that such and such ZODB objects constitute a path from root of a tree to some
of its leaf. After Track call the objects from the path and tree keys, that
are covered by leaf node, become tracked: from now-on ΔBtail will detect
and provide BTree-level changes caused by any change of tracked tree objects
or tracked keys. This guarantee can be provided because ΔBtail now knows
that such and such objects belong to a particular tree.

To manage knowledge which tree part is tracked ΔBtail uses PPTreeSubSet.
This data-structure represents so-called PP-connected set of tree nodes:
simply speaking it builds on some leafs and then includes parent(leaf),
parent(parent(leaf)), etc. In other words it's a "parent"-closure of the
leafs. The property of being PP-connected means that starting from any node
from such set, it is always possible to reach root node by traversing
.parent links, and that every intermediate node went-through during
traversal also belongs to the set.

A new Track request potentially grows tracked keys coverage. Due to this,
on a query, ΔBtail needs to recompute potentially whole vδT of the affected
tree. This recomputation is managed by "vδTSnapForTracked*" and "_rebuild"
functions and uses the same treediff algorithm, that Update is using, but
modulo PPTreeSubSet corresponding to δ key coverage. Update also potentially
needs to rebuild whole vδT history, not only append new δT, because a
change to tracked tree nodes can result in growth of tracked key coverage.

Queries are relatively straightforward code that work on vδT snapshot. The
main complexity, besides BTree-diff algorithm, lies in recomputing vδT when
set of tracked keys changes, and in handling that recomputation in such a way
that multiple Track and queries requests could be all served in parallel.

Concurrency

In order to allow multiple Track and queries requests to be served in
parallel ΔBtail employs special organization of vδT rebuild process where
complexity of concurrency is reduced to math on merging updates to vδT and
trackSet, and on key range lookup:

1. vδT is managed under read-copy-update (RCU) discipline: before making
   any vδT change the mutator atomically clones whole vδT and applies its
   change to the clone. This way a query, once it retrieves vδT snapshot,
   does not need to further synchronize with vδT mutators, and can rely on
   that retrieved vδT snapshot will remain immutable.

2. a Track request goes through 3 states: "new", "handle-in-progress" and
   "handled". At each state keys/nodes of the Track are maintained in:

   - ΔTtail.ktrackNew and .trackNew       for "new",
   - ΔTtail.krebuildJobs                  for "handle-in-progress", and
   - ΔBtail.trackSet                      for "handled".

   trackSet keeps nodes, and implicitly keys, from all handled Track
   requests. For all keys, covered by trackSet, vδT is fully computed.

   a new Track(keycov, path) is remembered in ktrackNew and trackNew to be
   further processed when a query should need keys from keycov. vδT is not
   yet providing data for keycov keys.

   when a Track request starts to be processed, its keys and nodes are moved
   from ktrackNew/trackNew into krebuildJobs. vδT is not yet providing data
   for requested-to-be-tracked keys.

   all trackSet, trackNew/ktrackNew and krebuildJobs are completely disjoint:

    trackSet ^ trackNew     = ø
    trackSet ^ krebuildJobs = ø
    trackNew ^ krebuildJobs = ø

3. when a query is served, it needs to retrieve vδT snapshot that takes
   related previous Track requests into account. Retrieving such snapshots
   is implemented in vδTSnapForTracked*() family of functions: there it
   checks ktrackNew/trackNew, and if those sets overlap with query's keys
   of interest, run vδT rebuild for keys queued in ktrackNew.

   the main part of that rebuild can be run without any locks, because it
   does not use nor modify any ΔBtail data, and for δ(vδT) it just computes
   a fresh full vδT build modulo retrieved ktrackNew. Only after that
   computation is complete, ΔBtail is locked again to quickly merge in
   δ(vδT) update back into vδT.

   This organization is based on the fact that

    vδT/(T₁∪T₂) = vδT/T₁ | vδT/T₂

     ( i.e. vδT computed for tracked set being union of T₁ and T₂ is the
       same as merge of vδT computed for tracked set T₁ and vδT computed
      for tracked set T₂ )

   and that

    trackSet | (δPP₁|δPP₂) = (trackSet|δPP₁) | (trackSet|δPP₂)

    ( i.e. tracking set updated for union of δPP₁ and δPP₂ is the same
      as union of tracking set updated with δPP₁ and tracking set updated
      with δPP₂ )

   these merge properties allow to run computation for δ(vδT) and δ(trackSet)
   independently and with ΔBtail unlocked, which in turn enables running
   several Track/queries in parallel.

4. while vδT rebuild is being run, krebuildJobs keeps corresponding keycov
   entry to indicate in-progress rebuild. Should a query need vδT for keys
   from that job, it first waits for corresponding job(s) to complete.

Explained rebuild organization allows non-overlapping queries/track-requests
to run simultaneously. (This property is essential to WCFS because otherwise
WCFS would not be able to serve several non-overlapping READ requests to one
file in parallel.)

--------

(*) implemented in treediff.go
(+) full database scan

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Some preliminary history:

kirr/wendelin.core@877e64a9    X wcfs: Fix tests to pass again
kirr/wendelin.core@c32055fc    X wcfs/xbtree: ΔBtail tests += ø -> Tree; Tree -> ø
kirr/wendelin.core@78f2f88b    X wcfs/xbtree: Fix treediff(a, ø)
kirr/wendelin.core@5324547c    X wcfs/xbtree: root(a) must stay in trackSet even after treediff(a,ø)
kirr/wendelin.core@f65f775b    X wcfs/xbtree: treediff(ø, b)
kirr/wendelin.core@c75b1c6f    X wcfs/xbtree: Start killing holeIdx
kirr/wendelin.core@0fa06cbd    X kadj must be taken into account as kadj^δZ
kirr/wendelin.core@ef5e5183    X treediff ret += δtkeycov
kirr/wendelin.core@f30826a6    X another bug in δtkeyconv computation
kirr/wendelin.core@0917380e    X wcfs: assert that keycov only grow
kirr/wendelin.core@502e05c2    X found why TestΔBTailAllStructs was not effective to find δtkeycov bugs
kirr/wendelin.core@450ba707    X Fix rebuild with ø @at2
kirr/wendelin.core@f60528c9    X ΔBtail.Clone had bug that it was aliasing klon and orig data
kirr/wendelin.core@9d20f8e8    X treediff: Fix BUG while computing AB coverage
kirr/wendelin.core@ddb28043    X rebuild: Don't return nil for empty ΔPPTreeSubSet - that leads to SIGSEGV
kirr/wendelin.core@324241eb    X rebuild: tests: Don't reflect.DeepEqual in inner loop
kirr/wendelin.core@8f6e2b1e    X rebuild: tests: Don't access ZODB in XGetδKV
kirr/wendelin.core@2c0b4793    X rebuild: tests: Don't access ZODB in xtrackKeys
kirr/wendelin.core@8f0e37f2    X rebuild: tests: Precompute kadj10·kadj21
kirr/wendelin.core@271d953d    X rebuild: tests: Move ΔBtail.Clone test out of hot inner loop into separate test
kirr/wendelin.core@a87cc6de    X rebuild: tests: Don't recompute trackSet(keys1R2) several times
kirr/wendelin.core@01433e96    X rebuild: tests: Don't compute keyCover in trackSet
kirr/wendelin.core@7371f9c5    X rebuild: tests: Inline _assertTrack
kirr/wendelin.core@3e9164b3    X rebuild: tests: Don't exercise keys from keys2 that already became tracked after Track(keys1) + Update
kirr/wendelin.core@e9c4b619    X rebuild: tests: Random testing
kirr/wendelin.core@d0fe680a    X δbtail += ForgetPast
kirr/wendelin.core@210e9b07    X Fix ΔBtail.SliceByRootRev (lo,hi] handling
kirr/wendelin.core@855ab4b8    X ΔBtail: Goodbye .KVAtTail
kirr/wendelin.core@2f5582e6    X ΔBtail: Tweak tests to run faster in normal mode
kirr/wendelin.core@cf352737    X random testing found another failing test for rebuild...
kirr/wendelin.core@7f7e34e0    X wcfs/xbtree: Fix update not to add duplicate extra point if rebuild  - called by Update - already added it
kirr/wendelin.core@6ad0052c    X ΔBtail.Track: No need to return error
kirr/wendelin.core@aafcacdf    X xbtree: GetAt test
kirr/wendelin.core@784a6761    X xbtree: Fix KAdj definition after treediff was reworked this summer to base decisions on node keycoverage instead of particular node keys
kirr/wendelin.core@0bb1c22e    X xbtree: Verify that ForgetPast clones vδT on trim
kirr/wendelin.core@a8945cbf    X Start reworking rebuild routines not to modify data inplace
kirr/wendelin.core@b74dda09    X Start switching Track from Track(key) to Track(keycov)
kirr/wendelin.core@dea85e87    X Switch GetAt to vδTSnapForTrackedKey
kirr/wendelin.core@aa0288ce    X Switch SliceByRootRev to vδTSnapForTracked
kirr/wendelin.core@c4366b14    X xbtree: tests: Also verify state of ΔTtail.ktrackNew
kirr/wendelin.core@b98706ad    X Track should be nop if keycov/path is already in krebuildJobs
kirr/wendelin.core@e141848a    X test.go  ↑ timeout  10m -> 20m
kirr/wendelin.core@423f77be    X wcfs: Goodby holeIdx
kirr/wendelin.core@37c2e806    X wcfs: Teach treediff to compute not only δtrack (set of nodes), but also δ for track-key coverage
kirr/wendelin.core@52c72dbb    X ΔBtail.rebuild started to work draftly
kirr/wendelin.core@c9f13fc7    X Get rebuild tests to run in a sane time; Add proper random-based testing for rebuild
kirr/wendelin.core@c7f1e3c9    X xbtree: Factor testing infrastructure bits into xbtree/xbtreetest
kirr/wendelin.core@7602c1f4    ΔBtail concurrency

This algorithm will be internally used by ΔBtail in the next patch.

The algorithm would be simple, if we would need to diff two trees
completely. However in ΔBtail only subpart of BTree nodes are tracked(*)
and the diff has to work modulo that tracking set.

No tests now because ΔBtail tests will cover treediff functionality as well.

Some preliminary history:

kirr/wendelin.core@78f2f88b    X wcfs/xbtree: Fix treediff(a, ø)
kirr/wendelin.core@5324547c    X wcfs/xbtree: root(a) must stay in trackSet even after treediff(a,ø)
kirr/wendelin.core@f65f775b    X wcfs/xbtree: treediff(ø, b)
kirr/wendelin.core@c75b1c6f    X wcfs/xbtree: Start killing holeIdx
kirr/wendelin.core@ef5e5183    X treediff ret += δtkeycov
kirr/wendelin.core@9d20f8e8    X treediff: Fix BUG while computing AB coverage
kirr/wendelin.core@ddb28043    X rebuild: Don't return nil for empty ΔPPTreeSubSet - that leads to SIGSEGV
kirr/wendelin.core@f68398c9    X wcfs: Move treediff into its own file

(*) because full BTree scan is needed to discover all of its nodes.

Quoting treediff documentation:

---- 8< ----

treediff provides diff for BTrees

Use δZConnectTracked + treediff to compute BTree-diff caused by δZ:

    δZConnectTracked(δZ, trackSet)                         -> δZTC, δtopsByRoot
    treediff(root, δtops, δZTC, trackSet, zconn{Old,New})  -> δT, δtrack, δtkeycov

δZConnectTracked computes BTree-connected closure of δZ modulo tracked set
and also returns δtopsByRoot to indicate which tree objects were changed and
in which subtree parts. With that information one can call treediff for each
changed root to compute BTree-diff and δ for trackSet itself.

BTree diff algorithm

diffT, diffB and δMerge constitute the diff algorithm implementation.
diff(A,B) works on pair of A and B whole key ranges splitted into regions
covered by tree nodes. The splitting represents current state of recursion
into corresponding tree. If a node in particular key range is Bucket, that
bucket contributes to δ- in case of A, and to δ+ in case of B. If a node in
particular key range is Tree, the algorithm may want to expand that tree
node into its children and to recourse into some of the children.

There are two phases:

- Phase 1 expands A top->down driven by δZTC, adds reached buckets to δ-,
  and queues key regions of those buckets to be processed on B.

- Phase 2 starts processing from queued key regions, expands them on B and
  adds reached buckets to δ+. Then it iterates to reach consistency in between
  A and B because processing buckets on B side may increase δ key coverage,
  and so corresponding key ranges has to be again processed on A. Which in
  turn may increase δ key coverage again, and needs to be processed on B side,
  etc...

The final δ is merge of δ- and δ+.

diffT has more detailed explanation of phase 1 and phase 2 logic.

This data structures will be used in ΔBtail to maintain sef of tracked
BTree nodes, and to represent δ to such set.

Some preliminary history:

kirr/wendelin.core@78f2f88b    X wcfs/xbtree: Fix treediff(a, ø)
kirr/wendelin.core@5324547c    X wcfs/xbtree: root(a) must stay in trackSet even after treediff(a,ø)
kirr/wendelin.core@f65f775b    X wcfs/xbtree: treediff(ø, b)
kirr/wendelin.core@66bc41ce    X Fix bug in PPTreeSubSet.Difference  - it was always leaving root node alive
kirr/wendelin.core@ddb28043    X rebuild: Don't return nil for empty ΔPPTreeSubSet - that leads to SIGSEGV
kirr/wendelin.core@a87cc6de    X rebuild: tests: Don't recompute trackSet(keys1R2) several times

Quoting PPTreeSubSet and ΔPPTreeSubSet documentation:

---- 8< ----

PPTreeSubSet represents PP-connected subset of tree node objects.

It is

    PP(xleafs)

where PP(node) maps node to {node, node.parent, node.parent,parent, ...} up
to top root from where the node is reached.

The nodes in the set are represented by their Oid.

Usually PPTreeSubSet is built as PP(some-leafs), but in general the starting
nodes are arbitrary. PPTreeSubSet can also have many root nodes, thus not
necessarily representing a subset of a single tree.

Usual set operations are provided: Union, Difference and Intersection.

Nodes can be added into the set via AddPath. Path is reverse operation - it
returns path to tree node given its oid.

Every node in the set comes with .parent pointer.

~~~~

ΔPPTreeSubSet represents a change to PPTreeSubSet.

It can be applied via PPTreeSubSet.ApplyΔ .

The result B of applying δ to A is:

    B = A.xDifference(δ.Del).xUnion(δ.Add)		(*)

(*) NOTE δ.Del and δ.Add might have their leafs starting from non-leaf nodes in A/B.
    This situation arises when δ represents a change in path to particular
    node, but that node itself does not change, for example:

           c*             c
          / \            /
        41*  42         41
         |    |         | \
        22   43        46  43
              |         |   |
             44        22  44

    Here nodes {c, 41} are changed, node 42 is unlinked, and node 46 is added.
    Nodes 43 and 44 stay unchanged.

        δ.Del = c-42-43   | c-41-22
        δ.Add = c-41-43   | c-41-46-22

    The second component with "-22" builds from leaf, but the first
    component with "-43" builds from non-leaf node.

        ΔnchildNonLeafs = {43: +1}

    Only complete result of applying all

        - xfixup(-1, ΔnchildNonLeafs)
        - δ.Del,
        - δ.Add, and
        - xfixup(+1, ΔnchildNonLeafs)

    produces correctly PP-connected set.

RangedMap is Key->VALUE map with adjacent keys mapped to the same value coalesced into Ranges.
RangedKeySet is set of Keys with adjacent keys coalesced into Ranges.

This data structures will be needed for ΔBtail.

For now the implementation is simple since it keeps whole map in a
linear slice because both RangedMap and RangedKeySet will be used in
ΔBtail to keep something proportional to δ of a change, which is assumed
to be small or medium most of the time.

Some preliminary history:

kirr/wendelin.core@6ea5920a    X xbtree: Less copy/garbage in RangedKeySet ops
kirr/wendelin.core@3ecacd99    X need to keep Value first so that sizeof(set-entry) = sizeof(KeyRange)
kirr/wendelin.core@a5b9b19b    X SetRange draftly works
kirr/wendelin.core@ed2de0de    X Tests for Get
kirr/wendelin.core@3b7b69e6    X fixes for empty set/range
kirr/wendelin.core@6972f999    X xbtree/blib: RangedMap, RangedSet += IntersectsRange, Intersection
kirr/wendelin.core@57be0126    X RangedMap - like RangedSet but for dict

Add treeenv.go that combines Treegen and client side access to ZODB with
committed trees as extension to testing.T . The environment allows to
easily see which tree update was committed, what is the difference in
terms of KV, what is the state of updated tree and state of pointed-to
ZBlk objects.

This will be used to test upcoming ΔBtail and ΔFtail.

Main functionality is in treeenv.go; the other added files are to
support that.

Some preliminary history:

kirr/wendelin.core@f07502fc    X xbtreetest: Teach T & Commit to automatically provide At in symbolic form
kirr/wendelin.core@0d62b05e    X Adjust to btree.VGet & friends signature change to include keycov in visit callback
kirr/wendelin.core@588a512a    X zdata: Switch SliceByFileRev not to clone Zinblk
kirr/wendelin.core@e9c4b619    X rebuild: tests: Random testing
kirr/wendelin.core@43090ac7    X tests: Factor-out tree-test-env into tTreeEnv
kirr/wendelin.core@d4a523b2    X δbtail: tests: Run much faster with live ZODB cache
kirr/wendelin.core@271d953d    X rebuild: tests: Move ΔBtail.Clone test out of hot inner loop into separate test
kirr/wendelin.core@c32055fc    X wcfs/xbtree: ΔBtail tests += ø -> Tree; Tree -> ø
kirr/wendelin.core@5324547c    X wcfs/xbtree: root(a) must stay in trackSet even after treediff(a,ø)
kirr/wendelin.core@8f6e2b1e    X rebuild: tests: Don't access ZODB in XGetδKV

Lacking generics we have set.go.in and instantiation for Set[int64],
set[string], Set[Oid] and Set[Tid] - that will be used in follow-up
patches.

The set.go.in itself is mostly a generalized copy from git-backup:

https://lab.nexedi.com/kirr/git-backup/blob/c9db60e8/set.go

treegen.go and treegen.py together provide a way

- to commit a particular BTree topology into ZODB, and
- to generate set of random tree topologies that all correspond to particular {k->v} dict.

this will be used in upcoming ΔBtail and ΔFtail tests.

See treegen.py documentation for details.

Some preliminary history:

kirr/wendelin.core@9eca74ec    X Teach AllStructs to emit topologies with values
kirr/wendelin.core@1b962f03    X Restructure: found bug that it was not marking objects as modified
kirr/wendelin.core@2139af2c    X treegen: Verify that tree actually saved to storage is what was requested
kirr/wendelin.core@b5e39d4a    X wcfs/treegen: allstructs: Do not keep all tree structures in memory
kirr/wendelin.core@e9c4b619    X rebuild: tests: Random testing
kirr/wendelin.core@c32055fc    X wcfs/xbtree: ΔBtail tests += ø -> Tree; Tree -> ø
kirr/wendelin.core@4300d88a    X wcfs/xbtreetest/treegen.py: Fix it on ZODB4

This will be the place to keep BTree-related utilities.
For now it provides only type aliases since Go lacks generics.

To handle invalidations, WCFS will need to detect changes to both ZBlk
objects and to ZBigFile.blktab BTree that is mapping file blocks to ZBlk
objects. And with BTree detecting changes is much more complex, because
when a BTree changes, it might be rebalanced, or keys migrated from one
tree/bucket node to another tree/bucket node. In other words a BTree
change might be not only a change to a {}key->value dictionary, but also
a change to BTree topology.

Because there are many BTree topologies that correspond to the same
{}key->value state, a change from kv₁ to kv₂, even if kv₁ and kv₂ are
close to each other, might be accompanied by a dramatic change to
topology of the tree. This creates a need for thoroughly testing the
BTree difference algorithm because many of BTree topologies changes are
tricky, and if a simple algorithm works on relatively stable topology
updates, it does not necessarily mean that that same algorithm will
continue to work correctly in the general case.

So, as a preparatory step, here comes xbtree.py package, that can be
used to inspect tree topologies, to create trees with specified topology
and to manipulate topology of an existing tree. This package will be
used in tests for upcoming ΔBtail.

For debugging, and also since those tests will involve both Go and
Python parts, it creates the need to be able to specify and exchange
topology of a tree via compact string. This package also defines so
called "topology encoding" to do so.

Some preliminar history:

kirr/wendelin.core@fb56193f    X fix metric to keep Z <- N order stable over key^
kirr/wendelin.core@809304d1    X "B:" indicates ø bucket with k&b, "B" - ø bucket with only keys
kirr/wendelin.core@9eca74ec    X Teach AllStructs to emit topologies with values
kirr/wendelin.core@1b962f03    X Restructure: found bug that it was not marking objects as modified
kirr/wendelin.core@9181c5d9    X Restructure; verify that it marks as changed only modifed nodes
kirr/wendelin.core@e9902c4a    X improve `xbtree topoview`

For the reference xbtree.py package documentation is quoted below.

---- 8< ----

Package xbtree provides utilities for inspecting/manipulating internal
structure of integer-keyed BTrees.

It will be primarily used to help verify ΔBTail in WCFS.

- `Tree` represents a tree node.
- `Bucket` represents a bucket node.
- `StructureOf` returns internal structure of ZODB BTree represented as Tree
  and Bucket nodes.
- `Restructure` reorganizes ZODB BTree instance according to specified topology
  structure.

- `AllStructs` generates all possible BTree topology structures with given keys.

Topology encoding
-----------------

Topology encoding provides way to represent structure of a Tree as path-like string.

TopoEncode converts Tree into its topology-encoded representation, while
TopoDecode decodes topology-encoded string back into Tree.

The following example illustrates topology encoding represented by string
"T3/T-T/B1-T5/B-B7,8,9":

      [ 3 ]             T3/         represents Tree([3])
       / \
     [ ] [ ]            T-T/        represents two empty Tree([])
      ↓   ↓
     |1|[ 5 ]           B1-T5/      represent Bucket([1]) and Tree([5])
         / \
        || |7|8|9|      B-B7,8,9    represents empty Bucket([]) and Bucket([7,8,9])

Topology encoding specification:

A Tree is encoded by level-order traversal, delimiting layers with "/".
Inside a layer Tree and Bucket nodes are signalled as

    "T<keys>"           ; Tree
    "B<keys>"           ; Bucket with only keys
    "B<keys+values>"    ; Bucket with keys and values

Keys are represented as ","-delimited list of integers. For example Tree
or Bucket with [1,3,5] keys are represented as

    "T1,3,5"        ; Tree([1,3,5])
    "B1,3,5"        ; Bucket([1,3,5])

Keys+values are represented as ","-delimited list of "<key>:<value>" pairs. For
example Bucket corresponding to {1:1, 2:4, 3:9} is represented as

    "B1:1,2:4,3:9"  ; Bucket([1,2,3], [1,4,9])

Empty keys+values are represented as ":" - an empty Bucket for key->value
mapping is represented as

    "B:"            ; Bucket([], [])

Nodes inside one layer are delimited with "-". For example a layer consisting
of an empty Tree, a Tree with [1,3] keys, and Bucket with [4,5] keys is
represented as

    "T-T1,3-B4,5"   ; layer with Tree([]), Tree([1,3]) and Bucket([4,5])

A layer consists of nodes that are followed by node-node links from upper layer
in left-to-right order.

Visualization
-------------

The following visualization utilities are provided to help understand BTrees
better:

- `topoview` displays BTree structure given its topology-encoded representation.
- `Tree.graphviz` returns Tree graph representation in dot language.

For WCFS to be efficient it will have to carefully preserve OS cache on
file invalidations. As preparatory step establish infrastructure for
verifying state of OS file cache and start asserting on OS cache state
in a couple of places.

See comments added to tFile constructor that describe how OS cache state
verification is setup.

Some preliminary history:

kirr/wendelin.core@8293025b    X Thoughts on how to avoid readahead touching pages of neighbour block
kirr/wendelin.core@3054e4a3    X not touching neighbour block works via setting MADV_RANDOM in last 1/4 of every block
kirr/wendelin.core@18362227    X #5 access still triggers read to #4 ?
kirr/wendelin.core@17dbf94e    X Provide mlock2 fallback for Ubuntu
kirr/wendelin.core@d134c0b9    X wcfs: test: try to live with only hard memlock limit adjusted
kirr/wendelin.core@c2423296    X Fix mlock2 build on Debian 8

Provide filesystem view of in-ZODB ZBigFiles, but do not implement support for
invalidations nor isolation protocol yet. In particular, because ZODB
invalidations are not yet handled, the filesystem does not update its data in
accordance with ZODB updates, and instead provides stale data view that
corresponds to the state of ZODB at the time when wcfs was mounted.

The main parts of this patch are:

- wcfs/wcfs.go is filesystem implementation itself together with overview.
- wcfs/__init__.py is python wrapper to spawn and interoperate with that filesystem.
- wcfs/wcfs_test.py is tests.

Some preliminary history:

fe7efb94    X start of wcfs
878b2787    X draft loading
d58c71e8    X don't overalign end by 1 blksize if end is already aligned
29c9f13d    X readBlk: Fix thinko in already case
59552328    X wcfs: Care to disable OS polling on us
c00d94c7    X workaround lack of exception chaining on Python2 with xdefer
0398e23d    X bytearray turned out to be copying data
7a837040    X print wcfs.py py-level traceback on SIGBUS (e.g. wcfs.go aborting due to bug/panic)
661b871f    X make sure tests don't get stuck even if wcfs gets killed -9 ...
2c043d29    X More effort to unmount failed wcfs.go
1ccc4478    X Use `with gil` + regular py code instead of PyGILState_Ensure/PyGILState_Release/PyRun_SimpleString
5dc9c791    X wcfs: Kill xdefer
91e9eba8    X wcfs: test: Register tFile to tDB early
a7138fef    X wcfs: mkdir /tmp/wcfs with sticky bit
1eec76d0    X wcfs: try to set sticky for /tmp/wcfs even if the directory already exists
c2c35851    X wcfs: tests: Factor-out waiting for a general condition to become true into waitfor
78f36993    X wcfs: test: Fix thinko in getting /sys/fs/fuse/connection/<X> for wcfs
bc9eb16f    X wcfs: tests: Don't use testmntpt everywhere
6dec74e7    X wcfs: tests: Split tDB into -> tDB + tWCFS
3a6bd764    X wcfs: tests: Run `fusermount -u` the second time if we had to kill wcfs
112720f3    X wcfs: tests: Print which files are still opened on wcfs if `fusermount -u` fails
bb40185b    X wcfs: Take $WENDELIN_CORE_WCFS_OPTIONS into account not only from under join
03a9ef33    X wcfs: Remove credentials from zurl when computing wcfs mountpoint
68ee5bdc    X wcfs: lsof tweaks
21671879    X wcfs: Teach entrypoint frontend to handle subcommands: serve, status, stop
b0642b80    X wcfs: Switch mountpoints from /tmp/wcfs/* to /dev/shm/*
b0ca031f    X wcfs: Teach join/serve to start successfully even after unclean wcfs shutdown
5bfa8cf8    X wcfs: Add start to spawn a Server that can be later stopped  (draft)
5fcec261    X wcfs: Run fusermount and friends with /bin:/usr/bin always on path
669d7a20    fixup! X wcfs: Run fusermount and friends with /bin:/usr/bin always on path
6b22f8c4    X wcfs: Teach start to start successfully even after unclean wcfs shutdown
15389db0    X wcfs: Tune _fuse_unmount to include `fusermount -u` error message into raised exception
153c002a    X wcfs: _fuse_unmount: Try first `kill -TERM` before `kill -QUIT` wcfs
3244f3a6    X wcfs: lsof +D misbehaves - don't use it
a126e709    X wcfs: Put client log into its own logger
ac303d1e    X wcfs: tests: -v  ->  show only wcfs.py logs verbosely
d671a9e9    X wcfs: Give more time to stop wcfs server

Add functionality to load objects from ZODB as saved by py wendelin.core.
Mostly straightforward code.
The main part is in zblk.go .

Contrary to python implementation, go can load ZBlk1's subobjects in
parallel, which, given scalable ZODB storage, can be significantly
faster compared to serially loading all ZData subobjects as py code
does.

TODO test wrt data saved by Python3.

Some preliminary history:

kirr/wendelin.core@878b2787 X draft loading
kirr/wendelin.core@bf9a7405 X No longer rely on ZODB cache invariant for invalidations
kirr/wendelin.core@0d62b05e X Adjust to btree.VGet & friends signature change to include keycov in visit callback
kirr/wendelin.core@b74dda09 X Start switching Track from Track(key) to Track(keycov)

Add initial stub for WCFS program and tests.
WCFS functionality will be added step-by-step in follow-up commits.

Some preliminary history:

0ae88a32       X .nxdtest: Verify Go bits with GOMAXPROCS=1,2,`nproc`
23528eb4       X wcfs: make it to use go modules for dependencies

In 6637d216 (lib/zodb: Add zstor_2zurl - way to convert a ZODB storage
into URL to access it) we added zstor_2zurl function to convert a ZODB
storage client object into an URL to access the storage. At that time
the function knew how to understand FileStorage only. Let's add support
for other storages that WCFS will need to support now.

NEO URI scheme matches the one currently used on ZODB/go side. It
semantically needs nexedi/neoppod!18
to be also applied to NEO/py side, but we do not care for now that that
patch is not merged (yet, or forever) because extracted ZURL is used
only with WCFS which uses NEO/go.

NEO support also depends on custom patch to remember SSL credentials on
NEO Client:

kirr/neo@a2f192cb

Some preliminary history:

kirr/wendelin.core@5cb39463    fixup! X wcfs/zeo started to work locally
kirr/wendelin.core@1cf3b228    X zstor_2zurl += NEO
kirr/wendelin.core@7f8fa32a    X lib/zodb: zstor_2zurl += NEO/SSL support
kirr/wendelin.core@e26524df    X wcfs, lib/zodb: DemoStorage support

Upcoming libwcfs (C++ part of WCFS client) will need to use virtmem code
and link to libvirtmem.

Manaully, because there is no automatic dependency tracking in
setuptools...

Dependency tracking is needed to avoid miscompilation after incremental
update under SlapOS/buildout/testnode/... when e.g. only .h was changed.

This is similar to e870781d (Top-level in-tree import redirector) but
for upcoming pyx modules.

Soon we are going to split virtmem code into its own DSO to which
bigfile extension will link. As plain setuptools does not support such
dynamic linking, we are going to use setuptools_dso instead. But more:
some of our upcoming extensions and DSOs will need to use Cython and C++
parts of Pygolang. Prepare that and use Extensions and DSO from
golang.pyx.build to support that right from the start.

Currently we have only one extension wendelin.bigfile._bigfile, but we
are going to add more both python extensions and non-python DSOs. Start
preparing to that by factoring-out common code.

lib/tests/testprog/zloadrace.py:90:1 'ZODB.FileStorage.FileStorage' imported but unused

This amends commit c37a989d.

Do what we can do without gdb and then tail to regular segmentation
fault. With core file gdb can still be used, but it is handy if we
already can get traceback of the crash into the log automatically.

TODO better use https://github.com/ianlancetaylor/libbacktrace because
backtrace_symbols often does not provide symbolic information.

We do not do this now because libbacktrace is not always automatically
installed.

This makes sure that those programs are always built afresh instead
being stuck at outdated build. This is needed because corresponding test .c
file includes many other .c files and we don't implement dependency tracking.

Else, e.g. after a failing test, that closed its storage and DB, but not
all Connections, another test, just by starting new transaction, would
invoke synchronization on that unclosed connection, which will try to
access closed storage and likely fail.

Fixes e.g. https://nexedijs.erp5.net/#/test_result_module/20210401-31B27B3D/5

Crash scenariou is the same as described in 5a5ed2c7 (tests: Force-close
ZODB connections in teardown, that testing code forgot to explicitly
close). Only now we try to isolate tests from each other not only for
different modules, but also for tests inside the same module.

The tests verify that there is no concurrency bugs around load,
Connection.open and invalidations. See e.g.

https://github.com/zopefoundation/ZODB/issues/290
https://github.com/zopefoundation/ZEO/issues/155

By including the tests into wendelin.core, we will have CI coverage for
all supported storages (FileStorage, ZEO, NEO), and for all supported
ZODB (currently ZODB4, ZODB4-wc2 and ZODB5).

ZEO5 is know to currently fail zloadrace.
However, even though ZODB#290 was fixed, ZEO5 turned out to also fail on zopenrace:

        def test_zodb_zopenrace():
            # exercises ZODB.Connection + particular storage implementation
    >       zopenrace.main()

    lib/tests/test_zodb.py:382:
    _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
    <decorator-gen-1>:2: in main
        ???
    ../../tools/go/pygolang/golang/__init__.py:103: in _
        return f(*argv, **kw)
    lib/tests/testprog/zopenrace.py:115: in main
        test(zstor)
    <decorator-gen-2>:2: in test
        ???
    ../../tools/go/pygolang/golang/__init__.py:103: in _
        return f(*argv, **kw)
    lib/tests/testprog/zopenrace.py:201: in test
        wg.wait()
    golang/_sync.pyx:246: in golang._sync.PyWorkGroup.wait
        ???
    golang/_sync.pyx:226: in golang._sync.PyWorkGroup.go.pyrunf
        ???
    lib/tests/testprog/zopenrace.py:165: in T1
        t1()
    _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

        def t1():
            transaction.begin()
            zconn = db.open()

            root = zconn.root()
            obj1 = root['obj1']
            obj2 = root['obj2']

            # obj1 - reload it from zstor
            # obj2 - get it from zconn cache
            obj1._p_invalidate()

            # both objects must have the same values
            i1 = obj1.i
            i2 = obj2.i
            if i1 != i2:
    >           raise AssertionError("T1: obj1.i (%d)  !=  obj2.i (%d)" % (i1, i2))
    E           AssertionError: T1: obj1.i (3)  !=  obj2.i (2)

    lib/tests/testprog/zopenrace.py:156: AssertionError

Previously if an assert or something failed in spawned thread, the main
thread was usually spinning indefinitely = tests hang. -> Switch all
threading places to use sync.WorkGroup and this way if a thread fails,
all other threads are canceled and the exception is reported back to
wg.wait in main thread.

Since we start to go this route, NotifyChannel is reworked to fully use
channels instead of busy-waiting.

setuptools_dso 2 started to emit those autogenerated files.
See https://github.com/mdavidsaver/setuptools_dso/pull/15 for details.

NEO 1.9 was released in 2018 and is outdated by now. NEO 1.12 is
currently the latest NEO release.

After switching to ZODB >= 4 in the previous commit, we can safely
require zodbtools, because there is now no conflict in between
ZODB3/ZODB eggs.

It's been a while since last ZODB3 3.10.7 release in 2016 and the last
commit in upstream ZODB3 repository (3.10 branch) is from 2017. The
world switched since then to ZODB4 and to ZODB5 after that.

We were still requiring ZODB3, because ZODB3 3.11 egg was just a
dependency on newer ZODB, ZEO, BTrees and persistent; and this way we
could be supporting all ZODB3.10.x and  ZODB4 and ZODB5 via ZODB3.11.

However upcoming Wendelin.core 2, for its proper working, needs MVCC
semantic as implemented in ZODB5. This forces us, even for ZODB4, to
backport non-trivial bits from ZODB5 (see [1]). Maintaining ZODB3
support at this point becomes non-practical, because, to our knowledge,
there is no wendelin.core user that plans to continue using ZODB3
without switching to at least ZODB4 in the near future.

So goodbye ZODB3. Even though ZODB still stays with us, it gives a
feeling similar to [2], because in 2014, when I was myself learning
ZODB, it was through ZODB3 - still at the time when all ZODB bits were
living together in one place.

[1] nexedi/ZODB!1
[2] https://lists.osuosl.org/pipermail/darcs-users/2008-September/014095.html

DB.close() does `del self.storage`.

https://github.com/zopefoundation/ZODB/blob/5.6.0-14-g0eae10cd0/src/ZODB/DB.py#L646

This way if DB was closed, but some conn(s) were not, it will crash in
teardown as e.g. below:

    _____________ ERROR at teardown of test_bigfile_zblk1_zdata_reuse ______________

        def teardown_module():
    >       testdb.teardown()

    bigfile/tests/test_filezodb.py:58:
    _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

    self = <wendelin.lib.testing.TestDB_ZEO object at 0x7fb9c0216350>

        def teardown(self):
            # close connections that test code forgot to close
            for connref, tb in self.connv:
                conn = connref()
                if conn is None:
                    continue
                if not conn.opened:
                    continue    # still alive, but closed
                print("W: testdb: teardown: %s left not closed by test code"
                      "; opened by:\n%s" % (conn, tb), file=sys.stderr)

                db = conn.db()
    >           stor = db.storage
    E           AttributeError: 'DB' object has no attribute 'storage'

    lib/testing.py:217: AttributeError

The fix is simple - don't use db.storage at all, because it is not actually used in that code.

Since ZBigFile keeps references to fileh objects that are created
through it it forms a file <=> fileh cycle that is not collected without
cyclic GC:

https://lab.nexedi.com/nexedi/wendelin.core/blob/v0.13-52-ga702d41/bigfile/file_zodb.py#L497
https://lab.nexedi.com/nexedi/wendelin.core/blob/v0.13-52-ga702d41/bigfile/file_zodb.py#L566-571

We did not noticed this leak until now because it is small, but with
upcoming wendelin.core 2 it is important to release a fileh, because
there is WCFS connection associated with fileh, and if fileh is not
released, that connection also stays alive, keeping on-WCFS resources
still being used, and preventing WCFS from being unmounted cleanly.

-> Add cyclic GC support to PyBigFile / PyBigFileH

NOTE: we still don't allow PyVMA <=> PyBigFileH cycles to be collected,
because fileh_close called from fileh.__del__ asserts that there are no
live mappings left. See added comments for details. There is no
known practical need to use such cycles, so this should be ok.

See also other patches on cyclic GC topic:

- 450ad804 (bigarray: ArrayRef support for BigArray)  // adds cyclic GC support for PyVMA
- d97641d2 (bigfile/py: Properly untrack PyVMA from GC before dealloc)

/proposed-for-review-on nexedi/wendelin.core!12

The logic in pyvma_traverse and pyvma_clear needs to be synchronized
with PyVMA deallocation. In the next patche we'll be amending this
logic, and it will help a reader to keep all those functions together.

For the reference: PyVMA support for cyclic GC was introduced in
450ad804 (bigarray: ArrayRef support for BigArray). See also d97641d2
(bigfile/py: Properly untrack PyVMA from GC before dealloc).

/proposed-for-review-on !12

Otherwise when /bin/sh is dash it fails with

    t/tfault-run: 35: test: on_pagefault: unexpected operator