##// END OF EJS Templates
wireproto: add streams to frame-based protocol...
wireproto: add streams to frame-based protocol Previously, the frame-based protocol was just a series of frames, with each frame associated with a request ID. In order to scale the protocol, we'll want to enable the use of compression. While it is possible to enable compression at the socket/pipe level, this has its disadvantages. The big one is it undermines the point of frames being standalone, atomic units that can be read and written: if you add compression above the framing protocol, you are back to having a stream-based protocol as opposed to something frame-based. So in order to preserve frames, compression needs to occur at the frame payload level. Compressing each frame's payload individually will limit compression ratios because the window size of the compressor will be limited by the max frame size, which is 32-64kb as currently defined. It will also add CPU overhead, as it is more efficient for compressors to operate on fewer, larger blocks of data than more, smaller blocks. So compressing each frame independently is out. This means we need to compress each frame's payload as if it is part of a larger stream. The simplest approach is to have 1 stream per connection. This could certainly work. However, it has disadvantages (documented below). We could also have 1 stream per RPC/command invocation. (This is the model HTTP/2 goes with.) This also has disadvantages. The main disadvantage to one global stream is that it has the very real potential to create CPU bottlenecks doing compression. Networks are only getting faster and the performance of single CPU cores has been relatively flat. Newer compression formats like zstandard offer better CPU cycle efficiency than predecessors like zlib. But it still all too common to saturate your CPU with compression overhead long before you saturate the network pipe. The main disadvantage with streams per request is that you can't reap the benefits of the compression context for multiple requests. For example, if you send 1000 RPC requests (or HTTP/2 requests for that matter), the response to each would have its own compression context. The overall size of the raw responses would be larger because compression contexts wouldn't be able to reference data from another request or response. The approach for streams as implemented in this commit is to support N streams per connection and for streams to potentially span requests and responses. As explained by the added internals docs, this facilitates servers and clients delegating independent streams and compression to independent threads / CPU cores. This helps alleviate the CPU bottleneck of compression. This design also allows compression contexts to be reused across requests/responses. This can result in improved compression ratios and less overhead for compressors and decompressors having to build new contexts. Another feature that was defined was the ability for individual frames within a stream to declare whether that individual frame's payload uses the content encoding (read: compression) defined by the stream. The idea here is that some servers may serve data from a combination of caches and dynamic resolution. Data coming from caches may be pre-compressed. We want to facilitate servers being able to essentially stream bytes from caches to the wire with minimal overhead. Being able to mix and match with frames are compressed within a stream enables these types of advanced server functionality. This commit defines the new streams mechanism. Basic code for supporting streams in frames has been added. But that code is seriously lacking and doesn't fully conform to the defined protocol. For example, we don't close any streams. And support for content encoding within streams is not yet implemented. The change was rather invasive and I didn't think it would be reasonable to implement the entire feature in a single commit. For the record, I would have loved to reuse an existing multiplexing protocol to build the new wire protocol on top of. However, I couldn't find a protocol that offers the performance and scaling characteristics that I desired. Namely, it should support multiple compression contexts to facilitate scaling out to multiple CPU cores and compression contexts should be able to live longer than single RPC requests. HTTP/2 *almost* fits the bill. But the semantics of HTTP message exchange state that streams can only live for a single request-response. We /could/ tunnel on top of HTTP/2 streams and frames with HEADER and DATA frames. But there's no guarantee that HTTP/2 libraries and proxies would allow us to use HTTP/2 streams and frames without the HTTP message exchange semantics defined in RFC 7540 Section 8. Other RPC protocols like gRPC tunnel are built on top of HTTP/2 and thus preserve its semantics of stream per RPC invocation. Even QUIC does this. We could attempt to invent a higher-level stream that spans HTTP/2 streams. But this would be violating HTTP/2 because there is no guarantee that HTTP/2 streams are routed to the same server. The best we can do - which is what this protocol does - is shoehorn all request and response data into a single HTTP message and create streams within. At that point, we've defined a Content-Type in HTTP parlance. It just so happens our media type can also work as a standalone, stream-based protocol, without leaning on HTTP or similar protocol. Differential Revision: https://phab.mercurial-scm.org/D2907

File last commit:

r36198:34e85044 default
r37304:9bfcbe4f default
Show More
graphmod.py
478 lines | 16.4 KiB | text/x-python | PythonLexer
# Revision graph generator for Mercurial
#
# Copyright 2008 Dirkjan Ochtman <dirkjan@ochtman.nl>
# Copyright 2007 Joel Rosdahl <joel@rosdahl.net>
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.
"""supports walking the history as DAGs suitable for graphical output
The most basic format we use is that of::
(id, type, data, [parentids])
The node and parent ids are arbitrary integers which identify a node in the
context of the graph returned. Type is a constant specifying the node type.
Data depends on type.
"""
from __future__ import absolute_import
from .node import nullrev
from . import (
dagop,
smartset,
util,
)
CHANGESET = 'C'
PARENT = 'P'
GRANDPARENT = 'G'
MISSINGPARENT = 'M'
# Style of line to draw. None signals a line that ends and is removed at this
# point. A number prefix means only the last N characters of the current block
# will use that style, the rest will use the PARENT style. Add a - sign
# (so making N negative) and all but the first N characters use that style.
EDGES = {PARENT: '|', GRANDPARENT: ':', MISSINGPARENT: None}
def dagwalker(repo, revs):
"""cset DAG generator yielding (id, CHANGESET, ctx, [parentinfo]) tuples
This generator function walks through revisions (which should be ordered
from bigger to lower). It returns a tuple for each node.
Each parentinfo entry is a tuple with (edgetype, parentid), where edgetype
is one of PARENT, GRANDPARENT or MISSINGPARENT. The node and parent ids
are arbitrary integers which identify a node in the context of the graph
returned.
"""
gpcache = {}
for rev in revs:
ctx = repo[rev]
# partition into parents in the rev set and missing parents, then
# augment the lists with markers, to inform graph drawing code about
# what kind of edge to draw between nodes.
pset = set(p.rev() for p in ctx.parents() if p.rev() in revs)
mpars = [p.rev() for p in ctx.parents()
if p.rev() != nullrev and p.rev() not in pset]
parents = [(PARENT, p) for p in sorted(pset)]
for mpar in mpars:
gp = gpcache.get(mpar)
if gp is None:
# precompute slow query as we know reachableroots() goes
# through all revs (issue4782)
if not isinstance(revs, smartset.baseset):
revs = smartset.baseset(revs)
gp = gpcache[mpar] = sorted(set(dagop.reachableroots(
repo, revs, [mpar])))
if not gp:
parents.append((MISSINGPARENT, mpar))
pset.add(mpar)
else:
parents.extend((GRANDPARENT, g) for g in gp if g not in pset)
pset.update(gp)
yield (ctx.rev(), CHANGESET, ctx, parents)
def nodes(repo, nodes):
"""cset DAG generator yielding (id, CHANGESET, ctx, [parentids]) tuples
This generator function walks the given nodes. It only returns parents
that are in nodes, too.
"""
include = set(nodes)
for node in nodes:
ctx = repo[node]
parents = set((PARENT, p.rev()) for p in ctx.parents()
if p.node() in include)
yield (ctx.rev(), CHANGESET, ctx, sorted(parents))
def colored(dag, repo):
"""annotates a DAG with colored edge information
For each DAG node this function emits tuples::
(id, type, data, (col, color), [(col, nextcol, color)])
with the following new elements:
- Tuple (col, color) with column and color index for the current node
- A list of tuples indicating the edges between the current node and its
parents.
"""
seen = []
colors = {}
newcolor = 1
config = {}
for key, val in repo.ui.configitems('graph'):
if '.' in key:
branch, setting = key.rsplit('.', 1)
# Validation
if setting == "width" and val.isdigit():
config.setdefault(branch, {})[setting] = int(val)
elif setting == "color" and val.isalnum():
config.setdefault(branch, {})[setting] = val
if config:
getconf = util.lrucachefunc(
lambda rev: config.get(repo[rev].branch(), {}))
else:
getconf = lambda rev: {}
for (cur, type, data, parents) in dag:
# Compute seen and next
if cur not in seen:
seen.append(cur) # new head
colors[cur] = newcolor
newcolor += 1
col = seen.index(cur)
color = colors.pop(cur)
next = seen[:]
# Add parents to next
addparents = [p for pt, p in parents if p not in next]
next[col:col + 1] = addparents
# Set colors for the parents
for i, p in enumerate(addparents):
if not i:
colors[p] = color
else:
colors[p] = newcolor
newcolor += 1
# Add edges to the graph
edges = []
for ecol, eid in enumerate(seen):
if eid in next:
bconf = getconf(eid)
edges.append((
ecol, next.index(eid), colors[eid],
bconf.get('width', -1),
bconf.get('color', '')))
elif eid == cur:
for ptype, p in parents:
bconf = getconf(p)
edges.append((
ecol, next.index(p), color,
bconf.get('width', -1),
bconf.get('color', '')))
# Yield and move on
yield (cur, type, data, (col, color), edges)
seen = next
def asciiedges(type, char, state, rev, parents):
"""adds edge info to changelog DAG walk suitable for ascii()"""
seen = state['seen']
if rev not in seen:
seen.append(rev)
nodeidx = seen.index(rev)
knownparents = []
newparents = []
for ptype, parent in parents:
if parent == rev:
# self reference (should only be seen in null rev)
continue
if parent in seen:
knownparents.append(parent)
else:
newparents.append(parent)
state['edges'][parent] = state['styles'].get(ptype, '|')
ncols = len(seen)
width = 1 + ncols * 2
nextseen = seen[:]
nextseen[nodeidx:nodeidx + 1] = newparents
edges = [(nodeidx, nextseen.index(p)) for p in knownparents]
seen[:] = nextseen
while len(newparents) > 2:
# ascii() only knows how to add or remove a single column between two
# calls. Nodes with more than two parents break this constraint so we
# introduce intermediate expansion lines to grow the active node list
# slowly.
edges.append((nodeidx, nodeidx))
edges.append((nodeidx, nodeidx + 1))
nmorecols = 1
width += 2
yield (type, char, width, (nodeidx, edges, ncols, nmorecols))
char = '\\'
nodeidx += 1
ncols += 1
edges = []
del newparents[0]
if len(newparents) > 0:
edges.append((nodeidx, nodeidx))
if len(newparents) > 1:
edges.append((nodeidx, nodeidx + 1))
nmorecols = len(nextseen) - ncols
if nmorecols > 0:
width += 2
# remove current node from edge characters, no longer needed
state['edges'].pop(rev, None)
yield (type, char, width, (nodeidx, edges, ncols, nmorecols))
def _fixlongrightedges(edges):
for (i, (start, end)) in enumerate(edges):
if end > start:
edges[i] = (start, end + 1)
def _getnodelineedgestail(
echars, idx, pidx, ncols, coldiff, pdiff, fix_tail):
if fix_tail and coldiff == pdiff and coldiff != 0:
# Still going in the same non-vertical direction.
if coldiff == -1:
start = max(idx + 1, pidx)
tail = echars[idx * 2:(start - 1) * 2]
tail.extend(["/", " "] * (ncols - start))
return tail
else:
return ["\\", " "] * (ncols - idx - 1)
else:
remainder = (ncols - idx - 1)
return echars[-(remainder * 2):] if remainder > 0 else []
def _drawedges(echars, edges, nodeline, interline):
for (start, end) in edges:
if start == end + 1:
interline[2 * end + 1] = "/"
elif start == end - 1:
interline[2 * start + 1] = "\\"
elif start == end:
interline[2 * start] = echars[2 * start]
else:
if 2 * end >= len(nodeline):
continue
nodeline[2 * end] = "+"
if start > end:
(start, end) = (end, start)
for i in range(2 * start + 1, 2 * end):
if nodeline[i] != "+":
nodeline[i] = "-"
def _getpaddingline(echars, idx, ncols, edges):
# all edges up to the current node
line = echars[:idx * 2]
# an edge for the current node, if there is one
if (idx, idx - 1) in edges or (idx, idx) in edges:
# (idx, idx - 1) (idx, idx)
# | | | | | | | |
# +---o | | o---+
# | | X | | X | |
# | |/ / | |/ /
# | | | | | |
line.extend(echars[idx * 2:(idx + 1) * 2])
else:
line.extend([' ', ' '])
# all edges to the right of the current node
remainder = ncols - idx - 1
if remainder > 0:
line.extend(echars[-(remainder * 2):])
return line
def _drawendinglines(lines, extra, edgemap, seen):
"""Draw ending lines for missing parent edges
None indicates an edge that ends at between this node and the next
Replace with a short line ending in ~ and add / lines to any edges to
the right.
"""
if None not in edgemap.values():
return
# Check for more edges to the right of our ending edges.
# We need enough space to draw adjustment lines for these.
edgechars = extra[::2]
while edgechars and edgechars[-1] is None:
edgechars.pop()
shift_size = max((edgechars.count(None) * 2) - 1, 0)
while len(lines) < 3 + shift_size:
lines.append(extra[:])
if shift_size:
empties = []
toshift = []
first_empty = extra.index(None)
for i, c in enumerate(extra[first_empty::2], first_empty // 2):
if c is None:
empties.append(i * 2)
else:
toshift.append(i * 2)
targets = list(range(first_empty, first_empty + len(toshift) * 2, 2))
positions = toshift[:]
for line in lines[-shift_size:]:
line[first_empty:] = [' '] * (len(line) - first_empty)
for i in range(len(positions)):
pos = positions[i] - 1
positions[i] = max(pos, targets[i])
line[pos] = '/' if pos > targets[i] else extra[toshift[i]]
map = {1: '|', 2: '~'}
for i, line in enumerate(lines):
if None not in line:
continue
line[:] = [c or map.get(i, ' ') for c in line]
# remove edges that ended
remove = [p for p, c in edgemap.items() if c is None]
for parent in remove:
del edgemap[parent]
seen.remove(parent)
def asciistate():
"""returns the initial value for the "state" argument to ascii()"""
return {
'seen': [],
'edges': {},
'lastcoldiff': 0,
'lastindex': 0,
'styles': EDGES.copy(),
'graphshorten': False,
}
def ascii(ui, state, type, char, text, coldata):
"""prints an ASCII graph of the DAG
takes the following arguments (one call per node in the graph):
- ui to write to
- Somewhere to keep the needed state in (init to asciistate())
- Column of the current node in the set of ongoing edges.
- Type indicator of node data, usually 'C' for changesets.
- Payload: (char, lines):
- Character to use as node's symbol.
- List of lines to display as the node's text.
- Edges; a list of (col, next_col) indicating the edges between
the current node and its parents.
- Number of columns (ongoing edges) in the current revision.
- The difference between the number of columns (ongoing edges)
in the next revision and the number of columns (ongoing edges)
in the current revision. That is: -1 means one column removed;
0 means no columns added or removed; 1 means one column added.
"""
idx, edges, ncols, coldiff = coldata
assert -2 < coldiff < 2
edgemap, seen = state['edges'], state['seen']
# Be tolerant of history issues; make sure we have at least ncols + coldiff
# elements to work with. See test-glog.t for broken history test cases.
echars = [c for p in seen for c in (edgemap.get(p, '|'), ' ')]
echars.extend(('|', ' ') * max(ncols + coldiff - len(seen), 0))
if coldiff == -1:
# Transform
#
# | | | | | |
# o | | into o---+
# |X / |/ /
# | | | |
_fixlongrightedges(edges)
# add_padding_line says whether to rewrite
#
# | | | | | | | |
# | o---+ into | o---+
# | / / | | | # <--- padding line
# o | | | / /
# o | |
add_padding_line = (len(text) > 2 and coldiff == -1 and
[x for (x, y) in edges if x + 1 < y])
# fix_nodeline_tail says whether to rewrite
#
# | | o | | | | o | |
# | | |/ / | | |/ /
# | o | | into | o / / # <--- fixed nodeline tail
# | |/ / | |/ /
# o | | o | |
fix_nodeline_tail = len(text) <= 2 and not add_padding_line
# nodeline is the line containing the node character (typically o)
nodeline = echars[:idx * 2]
nodeline.extend([char, " "])
nodeline.extend(
_getnodelineedgestail(
echars, idx, state['lastindex'], ncols, coldiff,
state['lastcoldiff'], fix_nodeline_tail))
# shift_interline is the line containing the non-vertical
# edges between this entry and the next
shift_interline = echars[:idx * 2]
for i in xrange(2 + coldiff):
shift_interline.append(' ')
count = ncols - idx - 1
if coldiff == -1:
for i in xrange(count):
shift_interline.extend(['/', ' '])
elif coldiff == 0:
shift_interline.extend(echars[(idx + 1) * 2:ncols * 2])
else:
for i in xrange(count):
shift_interline.extend(['\\', ' '])
# draw edges from the current node to its parents
_drawedges(echars, edges, nodeline, shift_interline)
# lines is the list of all graph lines to print
lines = [nodeline]
if add_padding_line:
lines.append(_getpaddingline(echars, idx, ncols, edges))
# If 'graphshorten' config, only draw shift_interline
# when there is any non vertical flow in graph.
if state['graphshorten']:
if any(c in '\/' for c in shift_interline if c):
lines.append(shift_interline)
# Else, no 'graphshorten' config so draw shift_interline.
else:
lines.append(shift_interline)
# make sure that there are as many graph lines as there are
# log strings
extra_interline = echars[:(ncols + coldiff) * 2]
if len(lines) < len(text):
while len(lines) < len(text):
lines.append(extra_interline[:])
_drawendinglines(lines, extra_interline, edgemap, seen)
while len(text) < len(lines):
text.append("")
if any(len(char) > 1 for char in edgemap.values()):
# limit drawing an edge to the first or last N lines of the current
# section the rest of the edge is drawn like a parent line.
parent = state['styles'][PARENT][-1:]
def _drawgp(char, i):
# should a grandparent character be drawn for this line?
if len(char) < 2:
return True
num = int(char[:-1])
# either skip first num lines or take last num lines, based on sign
return -num <= i if num < 0 else (len(lines) - i) <= num
for i, line in enumerate(lines):
line[:] = [c[-1:] if _drawgp(c, i) else parent for c in line]
edgemap.update(
(e, (c if len(c) < 2 else parent)) for e, c in edgemap.items())
# print lines
indentation_level = max(ncols, ncols + coldiff)
for (line, logstr) in zip(lines, text):
ln = "%-*s %s" % (2 * indentation_level, "".join(line), logstr)
ui.write(ln.rstrip() + '\n')
# ... and start over
state['lastcoldiff'] = coldiff
state['lastindex'] = idx