Merge remote-tracking branch 'origin/master' into 3948.static-setup

.circleci/config.yml

@@ -133,10 +133,10 @@ jobs:
     steps:
       - "checkout"
 
-      - run:
+      - run: &INSTALL_TOX
           name: "Install tox"
           command: |
-            pip install --user tox
+            pip install --user 'tox~=3.0'
 
       - run:
           name: "Static-ish code checks"
@@ -152,9 +152,7 @@ jobs:
       - "checkout"
 
       - run:
-          name: "Install tox"
+          <<: *INSTALL_TOX
-          command: |
-            pip install --user tox
 
       - run:
           name: "Make PyInstaller executable"

.circleci/populate-wheelhouse.sh

@@ -9,7 +9,7 @@ BASIC_DEPS="pip wheel setuptools setuptools_scm"
 
 # Python packages we need to support the test infrastructure.  *Not* packages
 # Tahoe-LAFS itself (implementation or test suite) need.
-TEST_DEPS="tox codecov"
+TEST_DEPS="tox~=3.0 codecov"
 
 # Python packages we need to generate test reports for CI infrastructure.
 # *Not* packages Tahoe-LAFS itself (implement or test suite) need.

.github/workflows/ci.yml

@@ -63,7 +63,7 @@ jobs:
             python-version: "pypy-3.7"
           - os: ubuntu-latest
             python-version: "pypy-3.8"
-    
+
     steps:
       # See https://github.com/actions/checkout. A fetch-depth of 0
       # fetches all tags and branches.
@@ -80,7 +80,7 @@ jobs:
 
       - name: Install Python packages
         run: |
-          pip install --upgrade codecov tox tox-gh-actions setuptools
+          pip install --upgrade codecov "tox<4" tox-gh-actions setuptools
           pip list
 
       - name: Display tool versions
@@ -199,7 +199,7 @@ jobs:
 
       - name: Install Python packages
         run: |
-          pip install --upgrade tox
+          pip install --upgrade "tox<4"
           pip list
 
       - name: Display tool versions
@@ -247,7 +247,7 @@ jobs:
 
       - name: Install Python packages
         run: |
-          pip install --upgrade tox
+          pip install --upgrade "tox<4"
           pip list
 
       - name: Display tool versions

newsfragments/3921.feature

@@ -0,0 +1,5 @@
+`tahoe run ...` will now exit when its stdin is closed.
+
+This facilitates subprocess management, specifically cleanup.
+When a parent process is running tahoe and exits without time to do "proper" cleanup at least the stdin descriptor will be closed.
+Subsequently "tahoe run" notices this and exits.

newsfragments/3939.bugfix

@@ -0,0 +1 @@
+Uploading immutables will now better use available bandwidth, which should allow for faster uploads in many cases.

setup.cfg

@@ -166,7 +166,7 @@ test =
     pyflakes == 2.2.0
     coverage ~= 5.0
     mock
-    tox
+    tox ~= 3.0
     pytest
     pytest-twisted
     hypothesis >= 3.6.1

src/allmydata/immutable/encode.py

@@ -262,6 +262,8 @@ class Encoder(object):
 
         d.addCallback(lambda res: self.finish_hashing())
 
+        # These calls have to happen in order; layout.py now requires writes to
+        # be appended to the data written so far.
         d.addCallback(lambda res:
                       self.send_crypttext_hash_tree_to_all_shareholders())
         d.addCallback(lambda res: self.send_all_block_hash_trees())

src/allmydata/immutable/layout.py

@@ -1,21 +1,18 @@
 """
 Ported to Python 3.
 """
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-from __future__ import unicode_literals
 
-from future.utils import PY2
+from __future__ import annotations
-if PY2:
-    from future.builtins import filter, map, zip, ascii, chr, hex, input, next, oct, open, pow, round, super, bytes, dict, list, object, range, str, max, min  # noqa: F401
 
 import struct
+from io import BytesIO
+
+from attrs import define, field
 from zope.interface import implementer
 from twisted.internet import defer
 from allmydata.interfaces import IStorageBucketWriter, IStorageBucketReader, \
      FileTooLargeError, HASH_SIZE
-from allmydata.util import mathutil, observer, pipeline, log
+from allmydata.util import mathutil, observer, log
 from allmydata.util.assertutil import precondition
 from allmydata.storage.server import si_b2a
 
@@ -107,19 +104,58 @@ def make_write_bucket_proxy(rref, server,
                                   num_share_hashes, uri_extension_size)
     return wbp
 
+
+@define
+class _WriteBuffer:
+    """
+    Queue up small writes to be written in a single batched larger write.
+    """
+    _batch_size: int
+    _to_write : BytesIO = field(factory=BytesIO)
+    _written_bytes : int = field(default=0)
+
+    def queue_write(self, data: bytes) -> bool:
+        """
+        Queue a write.  If the result is ``False``, no further action is needed
+        for now.  If the result is some ``True``, it's time to call ``flush()``
+        and do a real write.
+        """
+        self._to_write.write(data)
+        return self.get_queued_bytes() >= self._batch_size
+
+    def flush(self) -> tuple[int, bytes]:
+        """Return offset and data to be written."""
+        offset = self._written_bytes
+        data = self._to_write.getvalue()
+        self._written_bytes += len(data)
+        self._to_write = BytesIO()
+        return (offset, data)
+
+    def get_queued_bytes(self) -> int:
+        """Return number of queued, unwritten bytes."""
+        return self._to_write.tell()
+
+    def get_total_bytes(self) -> int:
+        """Return how many bytes were written or queued in total."""
+        return self._written_bytes + self.get_queued_bytes()
+
+
 @implementer(IStorageBucketWriter)
 class WriteBucketProxy(object):
+    """
+    Note: The various ``put_`` methods need to be called in the order in which the
+    bytes will get written.
+    """
     fieldsize = 4
     fieldstruct = ">L"
 
     def __init__(self, rref, server, data_size, block_size, num_segments,
-                 num_share_hashes, uri_extension_size, pipeline_size=50000):
+                 num_share_hashes, uri_extension_size, batch_size=1_000_000):
         self._rref = rref
         self._server = server
         self._data_size = data_size
         self._block_size = block_size
         self._num_segments = num_segments
-        self._written_bytes = 0
 
         effective_segments = mathutil.next_power_of_k(num_segments,2)
         self._segment_hash_size = (2*effective_segments - 1) * HASH_SIZE
@@ -130,11 +166,13 @@ class WriteBucketProxy(object):
 
         self._create_offsets(block_size, data_size)
 
-        # k=3, max_segment_size=128KiB gives us a typical segment of 43691
+        # With a ~1MB batch size, max upload speed is 1MB/(round-trip latency)
-        # bytes. Setting the default pipeline_size to 50KB lets us get two
+        # assuming the writing code waits for writes to finish, so 20MB/sec if
-        # segments onto the wire but not a third, which would keep the pipe
+        # latency is 50ms. In the US many people only have 1MB/sec upload speed
-        # filled.
+        # as of 2022 (standard Comcast). For further discussion of how one
-        self._pipeline = pipeline.Pipeline(pipeline_size)
+        # might set batch sizes see
+        # https://tahoe-lafs.org/trac/tahoe-lafs/ticket/3787#comment:1.
+        self._write_buffer = _WriteBuffer(batch_size)
 
     def get_allocated_size(self):
         return (self._offsets['uri_extension'] + self.fieldsize +
@@ -179,7 +217,7 @@ class WriteBucketProxy(object):
         return "<WriteBucketProxy for node %r>" % self._server.get_name()
 
     def put_header(self):
-        return self._write(0, self._offset_data)
+        return self._queue_write(0, self._offset_data)
 
     def put_block(self, segmentnum, data):
         offset = self._offsets['data'] + segmentnum * self._block_size
@@ -193,13 +231,13 @@ class WriteBucketProxy(object):
                                        (self._block_size *
                                         (self._num_segments - 1))),
                          len(data), self._block_size)
-        return self._write(offset, data)
+        return self._queue_write(offset, data)
 
     def put_crypttext_hashes(self, hashes):
         # plaintext_hash_tree precedes crypttext_hash_tree. It is not used, and
         # so is not explicitly written, but we need to write everything, so
         # fill it in with nulls.
-        d = self._write(self._offsets['plaintext_hash_tree'], b"\x00" * self._segment_hash_size)
+        d = self._queue_write(self._offsets['plaintext_hash_tree'], b"\x00" * self._segment_hash_size)
         d.addCallback(lambda _: self._really_put_crypttext_hashes(hashes))
         return d
 
@@ -212,7 +250,7 @@ class WriteBucketProxy(object):
         precondition(offset + len(data) <= self._offsets['block_hashes'],
                      offset, len(data), offset+len(data),
                      self._offsets['block_hashes'])
-        return self._write(offset, data)
+        return self._queue_write(offset, data)
 
     def put_block_hashes(self, blockhashes):
         offset = self._offsets['block_hashes']
@@ -223,7 +261,7 @@ class WriteBucketProxy(object):
         precondition(offset + len(data) <= self._offsets['share_hashes'],
                      offset, len(data), offset+len(data),
                      self._offsets['share_hashes'])
-        return self._write(offset, data)
+        return self._queue_write(offset, data)
 
     def put_share_hashes(self, sharehashes):
         # sharehashes is a list of (index, hash) tuples, so they get stored
@@ -237,29 +275,45 @@ class WriteBucketProxy(object):
         precondition(offset + len(data) <= self._offsets['uri_extension'],
                      offset, len(data), offset+len(data),
                      self._offsets['uri_extension'])
-        return self._write(offset, data)
+        return self._queue_write(offset, data)
 
     def put_uri_extension(self, data):
         offset = self._offsets['uri_extension']
         assert isinstance(data, bytes)
         precondition(len(data) == self._uri_extension_size)
         length = struct.pack(self.fieldstruct, len(data))
-        return self._write(offset, length+data)
+        return self._queue_write(offset, length+data)
 
-    def _write(self, offset, data):
+    def _queue_write(self, offset, data):
-        # use a Pipeline to pipeline several writes together. TODO: another
+        """
-        # speedup would be to coalesce small writes into a single call: this
+        This queues up small writes to be written in a single batched larger
-        # would reduce the foolscap CPU overhead per share, but wouldn't
+        write.
-        # reduce the number of round trips, so it might not be worth the
+
-        # effort.
+        Callers of this function are expected to queue the data in order, with
-        self._written_bytes += len(data)
+        no holes.  As such, the offset is technically unnecessary, but is used
-        return self._pipeline.add(len(data),
+        to check the inputs.  Possibly we should get rid of it.
-                                  self._rref.callRemote, "write", offset, data)
+        """
+        assert offset == self._write_buffer.get_total_bytes()
+        if self._write_buffer.queue_write(data):
+            return self._actually_write()
+        else:
+            return defer.succeed(False)
+
+    def _actually_write(self):
+        """Write data to the server."""
+        offset, data = self._write_buffer.flush()
+        return self._rref.callRemote("write", offset, data)
 
     def close(self):
-        assert self._written_bytes == self.get_allocated_size(), f"{self._written_bytes} != {self.get_allocated_size()}"
+        assert self._write_buffer.get_total_bytes() == self.get_allocated_size(), (
-        d = self._pipeline.add(0, self._rref.callRemote, "close")
+            f"{self._written_buffer.get_total_bytes_queued()} != {self.get_allocated_size()}"
-        d.addCallback(lambda ign: self._pipeline.flush())
+        )
+        if self._write_buffer.get_queued_bytes() > 0:
+            d = self._actually_write()
+        else:
+            # No data queued, don't send empty string write.
+            d = defer.succeed(True)
+        d.addCallback(lambda _: self._rref.callRemote("close"))
         return d
 
     def abort(self):
@@ -371,16 +425,16 @@ class ReadBucketProxy(object):
         self._fieldsize = fieldsize
         self._fieldstruct = fieldstruct
 
-        for field in ( 'data',
+        for field_name in ( 'data',
-                       'plaintext_hash_tree', # UNUSED
+                            'plaintext_hash_tree', # UNUSED
-                       'crypttext_hash_tree',
+                            'crypttext_hash_tree',
-                       'block_hashes',
+                            'block_hashes',
-                       'share_hashes',
+                            'share_hashes',
-                       'uri_extension',
+                            'uri_extension',
-                       ):
+                           ):
             offset = struct.unpack(fieldstruct, data[x:x+fieldsize])[0]
             x += fieldsize
-            self._offsets[field] = offset
+            self._offsets[field_name] = offset
         return self._offsets
 
     def _get_block_data(self, unused, blocknum, blocksize, thisblocksize):

src/allmydata/scripts/tahoe_run.py

@@ -21,7 +21,11 @@ from twisted.scripts import twistd
 from twisted.python import usage
 from twisted.python.filepath import FilePath
 from twisted.python.reflect import namedAny
-from twisted.internet.defer import maybeDeferred
+from twisted.python.failure import Failure
+from twisted.internet.defer import maybeDeferred, Deferred
+from twisted.internet.protocol import Protocol
+from twisted.internet.stdio import StandardIO
+from twisted.internet.error import ReactorNotRunning
 from twisted.application.service import Service
 
 from allmydata.scripts.default_nodedir import _default_nodedir
@@ -155,6 +159,8 @@ class DaemonizeTheRealService(Service, HookMixin):
 
     def startService(self):
 
+        from twisted.internet import reactor
+
         def start():
             node_to_instance = {
                 u"client": lambda: maybeDeferred(namedAny("allmydata.client.create_client"), self.basedir),
@@ -194,12 +200,14 @@ class DaemonizeTheRealService(Service, HookMixin):
 
             def created(srv):
                 srv.setServiceParent(self.parent)
+                # exiting on stdin-closed facilitates cleanup when run
+                # as a subprocess
+                on_stdin_close(reactor, reactor.stop)
             d.addCallback(created)
             d.addErrback(handle_config_error)
             d.addBoth(self._call_hook, 'running')
             return d
 
-        from twisted.internet import reactor
         reactor.callWhenRunning(start)
 
 
@@ -213,6 +221,46 @@ class DaemonizeTahoeNodePlugin(object):
         return DaemonizeTheRealService(self.nodetype, self.basedir, so)
 
 
+def on_stdin_close(reactor, fn):
+    """
+    Arrange for the function `fn` to run when our stdin closes
+    """
+    when_closed_d = Deferred()
+
+    class WhenClosed(Protocol):
+        """
+        Notify a Deferred when our connection is lost .. as this is passed
+        to twisted's StandardIO class, it is used to detect our parent
+        going away.
+        """
+
+        def connectionLost(self, reason):
+            when_closed_d.callback(None)
+
+    def on_close(arg):
+        try:
+            fn()
+        except ReactorNotRunning:
+            pass
+        except Exception:
+            # for our "exit" use-case failures will _mostly_ just be
+            # ReactorNotRunning (because we're already shutting down
+            # when our stdin closes) but no matter what "bad thing"
+            # happens we just want to ignore it .. although other
+            # errors might be interesting so we'll log those
+            print(Failure())
+        return arg
+
+    when_closed_d.addBoth(on_close)
+    # we don't need to do anything with this instance because it gets
+    # hooked into the reactor and thus remembered .. but we return it
+    # for Windows testing purposes.
+    return StandardIO(
+        proto=WhenClosed(),
+        reactor=reactor,
+    )
+
+
 def run(reactor, config, runApp=twistd.runApp):
     """
     Runs a Tahoe-LAFS node in the foreground.

src/allmydata/test/test_pipeline.py

@@ -1,198 +0,0 @@
-"""
-Tests for allmydata.util.pipeline.
-
-Ported to Python 3.
-"""
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-from __future__ import unicode_literals
-
-from future.utils import PY2
-if PY2:
-    from builtins import filter, map, zip, ascii, chr, hex, input, next, oct, open, pow, round, super, bytes, dict, list, object, range, str, max, min  # noqa: F401
-
-import gc
-
-from twisted.internet import defer
-from twisted.trial import unittest
-from twisted.python import log
-from twisted.python.failure import Failure
-
-from allmydata.util import pipeline
-
-
-class Pipeline(unittest.TestCase):
-    def pause(self, *args, **kwargs):
-        d = defer.Deferred()
-        self.calls.append( (d, args, kwargs) )
-        return d
-
-    def failUnlessCallsAre(self, expected):
-        #print(self.calls)
-        #print(expected)
-        self.failUnlessEqual(len(self.calls), len(expected), self.calls)
-        for i,c in enumerate(self.calls):
-            self.failUnlessEqual(c[1:], expected[i], str(i))
-
-    def test_basic(self):
-        self.calls = []
-        finished = []
-        p = pipeline.Pipeline(100)
-
-        d = p.flush() # fires immediately
-        d.addCallbacks(finished.append, log.err)
-        self.failUnlessEqual(len(finished), 1)
-        finished = []
-
-        d = p.add(10, self.pause, "one")
-        # the call should start right away, and our return Deferred should
-        # fire right away
-        d.addCallbacks(finished.append, log.err)
-        self.failUnlessEqual(len(finished), 1)
-        self.failUnlessEqual(finished[0], None)
-        self.failUnlessCallsAre([ ( ("one",) , {} ) ])
-        self.failUnlessEqual(p.gauge, 10)
-
-        # pipeline: [one]
-
-        finished = []
-        d = p.add(20, self.pause, "two", kw=2)
-        # pipeline: [one, two]
-
-        # the call and the Deferred should fire right away
-        d.addCallbacks(finished.append, log.err)
-        self.failUnlessEqual(len(finished), 1)
-        self.failUnlessEqual(finished[0], None)
-        self.failUnlessCallsAre([ ( ("one",) , {} ),
-                                  ( ("two",) , {"kw": 2} ),
-                                  ])
-        self.failUnlessEqual(p.gauge, 30)
-
-        self.calls[0][0].callback("one-result")
-        # pipeline: [two]
-        self.failUnlessEqual(p.gauge, 20)
-
-        finished = []
-        d = p.add(90, self.pause, "three", "posarg1")
-        # pipeline: [two, three]
-        flushed = []
-        fd = p.flush()
-        fd.addCallbacks(flushed.append, log.err)
-        self.failUnlessEqual(flushed, [])
-
-        # the call will be made right away, but the return Deferred will not,
-        # because the pipeline is now full.
-        d.addCallbacks(finished.append, log.err)
-        self.failUnlessEqual(len(finished), 0)
-        self.failUnlessCallsAre([ ( ("one",) , {} ),
-                                  ( ("two",) , {"kw": 2} ),
-                                  ( ("three", "posarg1"), {} ),
-                                  ])
-        self.failUnlessEqual(p.gauge, 110)
-
-        self.failUnlessRaises(pipeline.SingleFileError, p.add, 10, self.pause)
-
-        # retiring either call will unblock the pipeline, causing the #3
-        # Deferred to fire
-        self.calls[2][0].callback("three-result")
-        # pipeline: [two]
-
-        self.failUnlessEqual(len(finished), 1)
-        self.failUnlessEqual(finished[0], None)
-        self.failUnlessEqual(flushed, [])
-
-        # retiring call#2 will finally allow the flush() Deferred to fire
-        self.calls[1][0].callback("two-result")
-        self.failUnlessEqual(len(flushed), 1)
-
-    def test_errors(self):
-        self.calls = []
-        p = pipeline.Pipeline(100)
-
-        d1 = p.add(200, self.pause, "one")
-        d2 = p.flush()
-
-        finished = []
-        d1.addBoth(finished.append)
-        self.failUnlessEqual(finished, [])
-
-        flushed = []
-        d2.addBoth(flushed.append)
-        self.failUnlessEqual(flushed, [])
-
-        self.calls[0][0].errback(ValueError("oops"))
-
-        self.failUnlessEqual(len(finished), 1)
-        f = finished[0]
-        self.failUnless(isinstance(f, Failure))
-        self.failUnless(f.check(pipeline.PipelineError))
-        self.failUnlessIn("PipelineError", str(f.value))
-        self.failUnlessIn("ValueError", str(f.value))
-        r = repr(f.value)
-        self.failUnless("ValueError" in r, r)
-        f2 = f.value.error
-        self.failUnless(f2.check(ValueError))
-
-        self.failUnlessEqual(len(flushed), 1)
-        f = flushed[0]
-        self.failUnless(isinstance(f, Failure))
-        self.failUnless(f.check(pipeline.PipelineError))
-        f2 = f.value.error
-        self.failUnless(f2.check(ValueError))
-
-        # now that the pipeline is in the failed state, any new calls will
-        # fail immediately
-
-        d3 = p.add(20, self.pause, "two")
-
-        finished = []
-        d3.addBoth(finished.append)
-        self.failUnlessEqual(len(finished), 1)
-        f = finished[0]
-        self.failUnless(isinstance(f, Failure))
-        self.failUnless(f.check(pipeline.PipelineError))
-        r = repr(f.value)
-        self.failUnless("ValueError" in r, r)
-        f2 = f.value.error
-        self.failUnless(f2.check(ValueError))
-
-        d4 = p.flush()
-        flushed = []
-        d4.addBoth(flushed.append)
-        self.failUnlessEqual(len(flushed), 1)
-        f = flushed[0]
-        self.failUnless(isinstance(f, Failure))
-        self.failUnless(f.check(pipeline.PipelineError))
-        f2 = f.value.error
-        self.failUnless(f2.check(ValueError))
-
-    def test_errors2(self):
-        self.calls = []
-        p = pipeline.Pipeline(100)
-
-        d1 = p.add(10, self.pause, "one")
-        d2 = p.add(20, self.pause, "two")
-        d3 = p.add(30, self.pause, "three")
-        d4 = p.flush()
-
-        # one call fails, then the second one succeeds: make sure
-        # ExpandableDeferredList tolerates the second one
-
-        flushed = []
-        d4.addBoth(flushed.append)
-        self.failUnlessEqual(flushed, [])
-
-        self.calls[0][0].errback(ValueError("oops"))
-        self.failUnlessEqual(len(flushed), 1)
-        f = flushed[0]
-        self.failUnless(isinstance(f, Failure))
-        self.failUnless(f.check(pipeline.PipelineError))
-        f2 = f.value.error
-        self.failUnless(f2.check(ValueError))
-
-        self.calls[1][0].callback("two-result")
-        self.calls[2][0].errback(ValueError("three-error"))
-
-        del d1,d2,d3,d4
-        gc.collect()  # for PyPy

src/allmydata/test/test_runner.py

@@ -47,6 +47,9 @@ from twisted.internet.defer import (
     inlineCallbacks,
     DeferredList,
 )
+from twisted.internet.testing import (
+    MemoryReactorClock,
+)
 from twisted.python.filepath import FilePath
 from allmydata.util import fileutil, pollmixin
 from allmydata.util.encodingutil import unicode_to_argv
@@ -60,6 +63,9 @@ import allmydata
 from allmydata.scripts.runner import (
     parse_options,
 )
+from allmydata.scripts.tahoe_run import (
+    on_stdin_close,
+)
 
 from .common import (
     PIPE,
@@ -624,6 +630,64 @@ class RunNode(common_util.SignalMixin, unittest.TestCase, pollmixin.PollMixin):
         yield client_running
 
 
+def _simulate_windows_stdin_close(stdio):
+    """
+    on Unix we can just close all the readers, correctly "simulating"
+    a stdin close .. of course, Windows has to be difficult
+    """
+    stdio.writeConnectionLost()
+    stdio.readConnectionLost()
+
+
+class OnStdinCloseTests(SyncTestCase):
+    """
+    Tests for on_stdin_close
+    """
+
+    def test_close_called(self):
+        """
+        our on-close method is called when stdin closes
+        """
+        reactor = MemoryReactorClock()
+        called = []
+
+        def onclose():
+            called.append(True)
+        transport = on_stdin_close(reactor, onclose)
+        self.assertEqual(called, [])
+
+        if platform.isWindows():
+            _simulate_windows_stdin_close(transport)
+        else:
+            for reader in reactor.getReaders():
+                reader.loseConnection()
+            reactor.advance(1)  # ProcessReader does a callLater(0, ..)
+
+        self.assertEqual(called, [True])
+
+    def test_exception_ignored(self):
+        """
+        An exception from our on-close function is discarded.
+        """
+        reactor = MemoryReactorClock()
+        called = []
+
+        def onclose():
+            called.append(True)
+            raise RuntimeError("unexpected error")
+        transport = on_stdin_close(reactor, onclose)
+        self.assertEqual(called, [])
+
+        if platform.isWindows():
+            _simulate_windows_stdin_close(transport)
+        else:
+            for reader in reactor.getReaders():
+                reader.loseConnection()
+            reactor.advance(1)  # ProcessReader does a callLater(0, ..)
+
+        self.assertEqual(called, [True])
+
+
 class PidFileLocking(SyncTestCase):
     """
     Direct tests for allmydata.util.pid functions

src/allmydata/test/test_storage.py

@@ -3,14 +3,9 @@ Tests for allmydata.storage.
 
 Ported to Python 3.
 """
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-from __future__ import unicode_literals
 
-from future.utils import native_str, PY2, bytes_to_native_str, bchr
+from __future__ import annotations
-if PY2:
+from future.utils import native_str, bytes_to_native_str, bchr
-    from future.builtins import filter, map, zip, ascii, chr, hex, input, next, oct, open, pow, round, super, bytes, dict, list, object, range, str, max, min  # noqa: F401
 from six import ensure_str
 
 from io import (
@@ -59,7 +54,7 @@ from allmydata.storage.common import storage_index_to_dir, \
      si_b2a, si_a2b
 from allmydata.storage.lease import LeaseInfo
 from allmydata.immutable.layout import WriteBucketProxy, WriteBucketProxy_v2, \
-     ReadBucketProxy
+     ReadBucketProxy, _WriteBuffer
 from allmydata.mutable.layout import MDMFSlotWriteProxy, MDMFSlotReadProxy, \
                                      LayoutInvalid, MDMFSIGNABLEHEADER, \
                                      SIGNED_PREFIX, MDMFHEADER, \
@@ -3746,3 +3741,39 @@ class LeaseInfoTests(SyncTestCase):
             info.to_mutable_data(),
             HasLength(info.mutable_size()),
         )
+
+
+class WriteBufferTests(SyncTestCase):
+    """Tests for ``_WriteBuffer``."""
+
+    @given(
+        small_writes=strategies.lists(
+            strategies.binary(min_size=1, max_size=20),
+            min_size=10, max_size=20),
+        batch_size=strategies.integers(min_value=5, max_value=10)
+    )
+    def test_write_buffer(self, small_writes: list[bytes], batch_size: int):
+        """
+        ``_WriteBuffer`` coalesces small writes into bigger writes based on
+        the batch size.
+        """
+        wb = _WriteBuffer(batch_size)
+        result = b""
+        for data in small_writes:
+            should_flush = wb.queue_write(data)
+            if should_flush:
+                flushed_offset, flushed_data = wb.flush()
+                self.assertEqual(flushed_offset, len(result))
+                # The flushed data is in batch sizes, or closest approximation
+                # given queued inputs:
+                self.assertTrue(batch_size <= len(flushed_data) < batch_size + len(data))
+                result += flushed_data
+
+        # Final flush:
+        remaining_length = wb.get_queued_bytes()
+        flushed_offset, flushed_data = wb.flush()
+        self.assertEqual(remaining_length, len(flushed_data))
+        self.assertEqual(flushed_offset, len(result))
+        result += flushed_data
+
+        self.assertEqual(result, b"".join(small_writes))

src/allmydata/test/test_testing.py

@@ -9,18 +9,7 @@
 
 """
 Tests for the allmydata.testing helpers
-
-Ported to Python 3.
-
 """
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-from __future__ import unicode_literals
-
-from future.utils import PY2
-if PY2:
-    from future.builtins import filter, map, zip, ascii, chr, hex, input, next, oct, open, pow, round, super, bytes, dict, list, object, range, str, max, min  # noqa: F401
 
 from twisted.internet.defer import (
     inlineCallbacks,
@@ -56,10 +45,12 @@ from testtools.matchers import (
     IsInstance,
     MatchesStructure,
     AfterPreprocessing,
+    Contains,
 )
 from testtools.twistedsupport import (
     succeeded,
 )
+from twisted.web.http import GONE
 
 
 class FakeWebTest(SyncTestCase):
@@ -144,7 +135,8 @@ class FakeWebTest(SyncTestCase):
 
     def test_download_missing(self):
         """
-        Error if we download a capability that doesn't exist
+        The response to a request to download a capability that doesn't exist
+        is 410 (GONE).
         """
 
         http_client = create_tahoe_treq_client()
@@ -157,7 +149,11 @@ class FakeWebTest(SyncTestCase):
             resp,
             succeeded(
                 MatchesStructure(
-                    code=Equals(500)
+                    code=Equals(GONE),
+                    content=AfterPreprocessing(
+                        lambda m: m(),
+                        succeeded(Contains(b"No data for")),
+                    ),
                 )
             )
         )

src/allmydata/testing/web.py

@@ -6,20 +6,12 @@
 # This file is part of Tahoe-LAFS.
 #
 # See the docs/about.rst file for licensing information.
-"""Test-helpers for clients that use the WebUI.
-
-Ported to Python 3.
 
 """
-from __future__ import absolute_import
+Test-helpers for clients that use the WebUI.
-from __future__ import division
+"""
-from __future__ import print_function
-from __future__ import unicode_literals
-
-from future.utils import PY2
-if PY2:
-    from future.builtins import filter, map, zip, ascii, chr, hex, input, next, oct, open, pow, round, super, bytes, dict, list, object, range, str, max, min  # noqa: F401
 
+from __future__ import annotations
 
 import hashlib
 
@@ -54,6 +46,7 @@ import allmydata.uri
 from allmydata.util import (
     base32,
 )
+from ..util.dictutil import BytesKeyDict
 
 
 __all__ = (
@@ -147,7 +140,7 @@ class _FakeTahoeUriHandler(Resource, object):
 
     isLeaf = True
 
-    data = attr.ib(default=attr.Factory(dict))
+    data: BytesKeyDict = attr.ib(default=attr.Factory(BytesKeyDict))
     capability_generators = attr.ib(default=attr.Factory(dict))
 
     def _generate_capability(self, kind):
@@ -209,7 +202,7 @@ class _FakeTahoeUriHandler(Resource, object):
         capability = None
         for arg, value in uri.query:
             if arg == u"uri":
-                capability = value
+                capability = value.encode("utf-8")
         # it's legal to use the form "/uri/<capability>"
         if capability is None and request.postpath and request.postpath[0]:
             capability = request.postpath[0]
@@ -221,10 +214,9 @@ class _FakeTahoeUriHandler(Resource, object):
 
         # the user gave us a capability; if our Grid doesn't have any
         # data for it, that's an error.
-        capability = capability.encode('ascii')
         if capability not in self.data:
-            request.setResponseCode(http.BAD_REQUEST)
+            request.setResponseCode(http.GONE)
-            return u"No data for '{}'".format(capability.decode('ascii'))
+            return u"No data for '{}'".format(capability.decode('ascii')).encode("utf-8")
 
         return self.data[capability]
 

src/allmydata/util/dictutil.py

@@ -1,21 +1,6 @@
 """
 Tools to mess with dicts.
-
-Ported to Python 3.
 """
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-from __future__ import unicode_literals
-
-from future.utils import PY2
-if PY2:
-    # IMPORTANT: We deliberately don't import dict. The issue is that we're
-    # subclassing dict, so we'd end up exposing Python 3 dict APIs to lots of
-    # code that doesn't support it.
-    from builtins import filter, map, zip, ascii, chr, hex, input, next, oct, open, pow, round, super, bytes, list, object, range, str, max, min  # noqa: F401
-from six import ensure_str
-
 
 class DictOfSets(dict):
     def add(self, key, value):
@@ -104,7 +89,7 @@ def _make_enforcing_override(K, method_name):
             raise TypeError("{} must be of type {}".format(
                 repr(key), self.KEY_TYPE))
         return getattr(dict, method_name)(self, key, *args, **kwargs)
-    f.__name__ = ensure_str(method_name)
+    f.__name__ = method_name
     setattr(K, method_name, f)
 
 for _method_name in ["__setitem__", "__getitem__", "setdefault", "get",
@@ -113,18 +98,13 @@ for _method_name in ["__setitem__", "__getitem__", "setdefault", "get",
 del _method_name
 
 
-if PY2:
+class BytesKeyDict(_TypedKeyDict):
-    # No need for enforcement, can use either bytes or unicode as keys and it's
+    """Keys should be bytes."""
-    # fine.
-    BytesKeyDict = UnicodeKeyDict = dict
-else:
-    class BytesKeyDict(_TypedKeyDict):
-        """Keys should be bytes."""
 
-        KEY_TYPE = bytes
+    KEY_TYPE = bytes
 
 
-    class UnicodeKeyDict(_TypedKeyDict):
+class UnicodeKeyDict(_TypedKeyDict):
-        """Keys should be unicode strings."""
+    """Keys should be unicode strings."""
 
-        KEY_TYPE = str
+    KEY_TYPE = str

src/allmydata/util/pipeline.py

@@ -1,149 +0,0 @@
-"""
-A pipeline of Deferreds.
-
-Ported to Python 3.
-"""
-
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-from __future__ import unicode_literals
-
-from future.utils import PY2
-if PY2:
-    from builtins import filter, map, zip, ascii, chr, hex, input, next, oct, open, pow, round, super, bytes, dict, list, object, range, str, max, min  # noqa: F401
-
-from twisted.internet import defer
-from twisted.python.failure import Failure
-from twisted.python import log
-from allmydata.util.assertutil import precondition
-
-
-class PipelineError(Exception):
-    """One of the pipelined messages returned an error. The received Failure
-    object is stored in my .error attribute."""
-    def __init__(self, error):
-        self.error = error
-
-    def __repr__(self):
-        return "<PipelineError error=(%r)>" % (self.error,)
-    def __str__(self):
-        return "<PipelineError error=(%s)>" % (self.error,)
-
-class SingleFileError(Exception):
-    """You are not permitted to add a job to a full pipeline."""
-
-
-class ExpandableDeferredList(defer.Deferred, object):
-    # like DeferredList(fireOnOneErrback=True) with a built-in
-    # gatherResults(), but you can add new Deferreds until you close it. This
-    # gives you a chance to add don't-complain-about-unhandled-error errbacks
-    # immediately after attachment, regardless of whether you actually end up
-    # wanting the list or not.
-    def __init__(self):
-        defer.Deferred.__init__(self)
-        self.resultsReceived = 0
-        self.resultList = []
-        self.failure = None
-        self.closed = False
-
-    def addDeferred(self, d):
-        precondition(not self.closed, "don't call addDeferred() on a closed ExpandableDeferredList")
-        index = len(self.resultList)
-        self.resultList.append(None)
-        d.addCallbacks(self._cbDeferred, self._ebDeferred,
-                       callbackArgs=(index,))
-        return d
-
-    def close(self):
-        self.closed = True
-        self.checkForFinished()
-
-    def checkForFinished(self):
-        if not self.closed:
-            return
-        if self.called:
-            return
-        if self.failure:
-            self.errback(self.failure)
-        elif self.resultsReceived == len(self.resultList):
-            self.callback(self.resultList)
-
-    def _cbDeferred(self, res, index):
-        self.resultList[index] = res
-        self.resultsReceived += 1
-        self.checkForFinished()
-        return res
-
-    def _ebDeferred(self, f):
-        self.failure = f
-        self.checkForFinished()
-        return f
-
-
-class Pipeline(object):
-    """I manage a size-limited pipeline of Deferred operations, usually
-    callRemote() messages."""
-
-    def __init__(self, capacity):
-        self.capacity = capacity # how full we can be
-        self.gauge = 0 # how full we are
-        self.failure = None
-        self.waiting = [] # callers of add() who are blocked
-        self.unflushed = ExpandableDeferredList()
-
-    def add(self, _size, _func, *args, **kwargs):
-        # We promise that all the Deferreds we return will fire in the order
-        # they were returned. To make it easier to keep this promise, we
-        # prohibit multiple outstanding calls to add() .
-        if self.waiting:
-            raise SingleFileError
-        if self.failure:
-            return defer.fail(self.failure)
-        self.gauge += _size
-        fd = defer.maybeDeferred(_func, *args, **kwargs)
-        fd.addBoth(self._call_finished, _size)
-        self.unflushed.addDeferred(fd)
-        fd.addErrback(self._eat_pipeline_errors)
-        fd.addErrback(log.err, "_eat_pipeline_errors didn't eat it")
-        if self.gauge < self.capacity:
-            return defer.succeed(None)
-        d = defer.Deferred()
-        self.waiting.append(d)
-        return d
-
-    def flush(self):
-        if self.failure:
-            return defer.fail(self.failure)
-        d, self.unflushed = self.unflushed, ExpandableDeferredList()
-        d.close()
-        d.addErrback(self._flushed_error)
-        return d
-
-    def _flushed_error(self, f):
-        precondition(self.failure) # should have been set by _call_finished
-        return self.failure
-
-    def _call_finished(self, res, size):
-        self.gauge -= size
-        if isinstance(res, Failure):
-            res = Failure(PipelineError(res))
-            if not self.failure:
-                self.failure = res
-        if self.failure:
-            while self.waiting:
-                d = self.waiting.pop(0)
-                d.errback(self.failure)
-        else:
-            while self.waiting and (self.gauge < self.capacity):
-                d = self.waiting.pop(0)
-                d.callback(None)
-                # the d.callback() might trigger a new call to add(), which
-                # will raise our gauge and might cause the pipeline to be
-                # filled. So the while() loop gets a chance to tell the
-                # caller to stop.
-        return res
-
-    def _eat_pipeline_errors(self, f):
-        f.trap(PipelineError)
-        return None

tox.ini

@@ -86,7 +86,6 @@ commands =
          coverage: python -b -m coverage run -m twisted.trial {env:TAHOE_LAFS_TRIAL_ARGS:--rterrors --reporter=timing} {posargs:{env:TEST_SUITE}}
          coverage: coverage combine
          coverage: coverage xml
-         coverage: coverage report
 
 [testenv:integration]
 basepython = python3
@@ -99,7 +98,6 @@ commands =
          # NOTE: 'run with "py.test --keep-tempdir -s -v integration/" to debug failures'
          py.test --timeout=1800 --coverage -s -v {posargs:integration}
          coverage combine
-         coverage report
 
 
 [testenv:codechecks]