hgweb: use separate repo instances per thread

Before this change, multiple threads/requests could share a
localrepository instance. This meant that all of localrepository needed
to be thread safe. Many bugs have been reported telling us that
localrepository isn't actually thread safe.

While making localrepository thread safe is a noble cause, it is a lot
of work. And there is little gain from doing so. Due to Python's GIL,
only 1 thread may be processing Python code at a time. The benefits
to multi-threaded servers are marginal.

Thread safety would be a lot of work for little gain. So, we're not
going to even attempt it.

This patch establishes a pool of repos in hgweb. When a request arrives,
we obtain the most recently used repository from the pool or create a
new one if none is available. When the request has finished, we put that
repo back in the pool.

We start with a pool size of 1. For servers using a single thread, the
pool will only ever be of size 1. For multi-threaded servers, the pool
size will grow to the max number of simultaneous requests the server
processes.

No logic for pruning the pool has been implemented. We assume server
operators either limit the number of threads to something they can
handle or restart the Mercurial process after a certain amount of
requests or time has passed.

mercurial/hgweb/hgweb_mod.py

@@ -1,411 +1,437
 # hgweb/hgweb_mod.py - Web interface for a repository.
 #
 # Copyright 21 May 2005 - (c) 2005 Jake Edge <[e-mail]>
 # Copyright 2005-2007 Matt Mackall <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
+import contextlib
 import os
 from mercurial import ui, hg, hook, error, encoding, templater, util, repoview
 from mercurial.templatefilters import websub
 from common import get_stat, ErrorResponse, permhooks, caching
 from common import HTTP_OK, HTTP_NOT_MODIFIED, HTTP_BAD_REQUEST
 from common import HTTP_NOT_FOUND, HTTP_SERVER_ERROR
 from request import wsgirequest
 import webcommands, protocol, webutil
 
 perms = {
     'changegroup': 'pull',
     'changegroupsubset': 'pull',
     'getbundle': 'pull',
     'stream_out': 'pull',
     'listkeys': 'pull',
     'unbundle': 'push',
     'pushkey': 'push',
 }
 
 def makebreadcrumb(url, prefix=''):
     '''Return a 'URL breadcrumb' list
 
     A 'URL breadcrumb' is a list of URL-name pairs,
     corresponding to each of the path items on a URL.
     This can be used to create path navigation entries.
     '''
     if url.endswith('/'):
         url = url[:-1]
     if prefix:
         url = '/' + prefix + url
     relpath = url
     if relpath.startswith('/'):
         relpath = relpath[1:]
 
     breadcrumb = []
     urlel = url
     pathitems = [''] + relpath.split('/')
     for pathel in reversed(pathitems):
         if not pathel or not urlel:
             break
         breadcrumb.append({'url': urlel, 'name': pathel})
         urlel = os.path.dirname(urlel)
     return reversed(breadcrumb)
 
 class requestcontext(object):
     """Holds state/context for an individual request.
 
     Servers can be multi-threaded. Holding state on the WSGI application
     is prone to race conditions. Instances of this class exist to hold
     mutable and race-free state for requests.
     """
     def __init__(self, app, repo):
         self.repo = repo
         self.reponame = app.reponame
 
         self.archives = ('zip', 'gz', 'bz2')
 
         self.maxchanges = self.configint('web', 'maxchanges', 10)
         self.stripecount = self.configint('web', 'stripes', 1)
         self.maxshortchanges = self.configint('web', 'maxshortchanges', 60)
         self.maxfiles = self.configint('web', 'maxfiles', 10)
         self.allowpull = self.configbool('web', 'allowpull', True)
 
         # we use untrusted=False to prevent a repo owner from using
         # web.templates in .hg/hgrc to get access to any file readable
         # by the user running the CGI script
         self.templatepath = self.config('web', 'templates', untrusted=False)
 
         # This object is more expensive to build than simple config values.
         # It is shared across requests. The app will replace the object
         # if it is updated. Since this is a reference and nothing should
         # modify the underlying object, it should be constant for the lifetime
         # of the request.
         self.websubtable = app.websubtable
 
     # Trust the settings from the .hg/hgrc files by default.
     def config(self, section, name, default=None, untrusted=True):
         return self.repo.ui.config(section, name, default,
                                    untrusted=untrusted)
 
     def configbool(self, section, name, default=False, untrusted=True):
         return self.repo.ui.configbool(section, name, default,
                                        untrusted=untrusted)
 
     def configint(self, section, name, default=None, untrusted=True):
         return self.repo.ui.configint(section, name, default,
                                       untrusted=untrusted)
 
     def configlist(self, section, name, default=None, untrusted=True):
         return self.repo.ui.configlist(section, name, default,
                                        untrusted=untrusted)
 
     archivespecs = {
         'bz2': ('application/x-bzip2', 'tbz2', '.tar.bz2', None),
         'gz': ('application/x-gzip', 'tgz', '.tar.gz', None),
         'zip': ('application/zip', 'zip', '.zip', None),
     }
 
     def archivelist(self, nodeid):
         allowed = self.configlist('web', 'allow_archive')
         for typ, spec in self.archivespecs.iteritems():
             if typ in allowed or self.configbool('web', 'allow%s' % typ):
                 yield {'type': typ, 'extension': spec[2], 'node': nodeid}
 
     def templater(self, req):
         # determine scheme, port and server name
         # this is needed to create absolute urls
 
         proto = req.env.get('wsgi.url_scheme')
         if proto == 'https':
             proto = 'https'
             default_port = '443'
         else:
             proto = 'http'
             default_port = '80'
 
         port = req.env['SERVER_PORT']
         port = port != default_port and (':' + port) or ''
         urlbase = '%s://%s%s' % (proto, req.env['SERVER_NAME'], port)
         logourl = self.config('web', 'logourl', 'http://mercurial.selenic.com/')
         logoimg = self.config('web', 'logoimg', 'hglogo.png')
         staticurl = self.config('web', 'staticurl') or req.url + 'static/'
         if not staticurl.endswith('/'):
             staticurl += '/'
 
         # some functions for the templater
 
         def motd(**map):
             yield self.config('web', 'motd', '')
 
         # figure out which style to use
 
         vars = {}
         styles = (
             req.form.get('style', [None])[0],
             self.config('web', 'style'),
             'paper',
         )
         style, mapfile = templater.stylemap(styles, self.templatepath)
         if style == styles[0]:
             vars['style'] = style
 
         start = req.url[-1] == '?' and '&' or '?'
         sessionvars = webutil.sessionvars(vars, start)
 
         if not self.reponame:
             self.reponame = (self.config('web', 'name')
                              or req.env.get('REPO_NAME')
                              or req.url.strip('/') or self.repo.root)
 
         def websubfilter(text):
             return websub(text, self.websubtable)
 
         # create the templater
 
         tmpl = templater.templater(mapfile,
                                    filters={'websub': websubfilter},
                                    defaults={'url': req.url,
                                              'logourl': logourl,
                                              'logoimg': logoimg,
                                              'staticurl': staticurl,
                                              'urlbase': urlbase,
                                              'repo': self.reponame,
                                              'encoding': encoding.encoding,
                                              'motd': motd,
                                              'sessionvars': sessionvars,
                                              'pathdef': makebreadcrumb(req.url),
                                              'style': style,
                                             })
         return tmpl
 
 
 class hgweb(object):
     """HTTP server for individual repositories.
 
     Instances of this class serve HTTP responses for a particular
     repository.
 
     Instances are typically used as WSGI applications.
 
     Some servers are multi-threaded. On these servers, there may
     be multiple active threads inside __call__.
     """
     def __init__(self, repo, name=None, baseui=None):
         if isinstance(repo, str):
             if baseui:
                 u = baseui.copy()
             else:
                 u = ui.ui()
             r = hg.repository(u, repo)
         else:
             # we trust caller to give us a private copy
             r = repo
 
         r.ui.setconfig('ui', 'report_untrusted', 'off', 'hgweb')
         r.baseui.setconfig('ui', 'report_untrusted', 'off', 'hgweb')
         r.ui.setconfig('ui', 'nontty', 'true', 'hgweb')
         r.baseui.setconfig('ui', 'nontty', 'true', 'hgweb')
         # displaying bundling progress bar while serving feel wrong and may
         # break some wsgi implementation.
         r.ui.setconfig('progress', 'disable', 'true', 'hgweb')
         r.baseui.setconfig('progress', 'disable', 'true', 'hgweb')
-        self._repo = hg.cachedlocalrepo(self._webifyrepo(r))
+        self._repos = [hg.cachedlocalrepo(self._webifyrepo(r))]
+        self._lastrepo = self._repos[0]
         hook.redirect(True)
         self.reponame = name
 
     def _webifyrepo(self, repo):
         repo = getwebview(repo)
         self.websubtable = webutil.getwebsubs(repo)
         return repo
 
-    def _getrepo(self):
-        r, created = self._repo.fetch()
+    @contextlib.contextmanager
+    def _obtainrepo(self):
+        """Obtain a repo unique to the caller.
+
+        Internally we maintain a stack of cachedlocalrepo instances
+        to be handed out. If one is available, we pop it and return it,
+        ensuring it is up to date in the process. If one is not available,
+        we clone the most recently used repo instance and return it.
+
+        It is currently possible for the stack to grow without bounds
+        if the server allows infinite threads. However, servers should
+        have a thread limit, thus establishing our limit.
+        """
+        if self._repos:
+            cached = self._repos.pop()
+            r, created = cached.fetch()
         if created:
             r = self._webifyrepo(r)
+        else:
+            cached = self._lastrepo.copy()
+            r, created = cached.fetch()
 
-        self.mtime = self._repo.mtime
-        return r
+        self._lastrepo = cached
+        self.mtime = cached.mtime
+        try:
+            yield r
+        finally:
+            self._repos.append(cached)
 
     def run(self):
         """Start a server from CGI environment.
 
         Modern servers should be using WSGI and should avoid this
         method, if possible.
         """
         if not os.environ.get('GATEWAY_INTERFACE', '').startswith("CGI/1."):
             raise RuntimeError("This function is only intended to be "
                                "called while running as a CGI script.")
         import mercurial.hgweb.wsgicgi as wsgicgi
         wsgicgi.launch(self)
 
     def __call__(self, env, respond):
         """Run the WSGI application.
 
         This may be called by multiple threads.
         """
         req = wsgirequest(env, respond)
         return self.run_wsgi(req)
 
     def run_wsgi(self, req):
         """Internal method to run the WSGI application.
 
         This is typically only called by Mercurial. External consumers
         should be using instances of this class as the WSGI application.
         """
-        repo = self._getrepo()
+        with self._obtainrepo() as repo:
+            return self._runwsgi(req, repo)
+
+    def _runwsgi(self, req, repo):
         rctx = requestcontext(self, repo)
 
         # This state is global across all threads.
         encoding.encoding = rctx.config('web', 'encoding', encoding.encoding)
         rctx.repo.ui.environ = req.env
 
         # work with CGI variables to create coherent structure
         # use SCRIPT_NAME, PATH_INFO and QUERY_STRING as well as our REPO_NAME
 
         req.url = req.env['SCRIPT_NAME']
         if not req.url.endswith('/'):
             req.url += '/'
         if 'REPO_NAME' in req.env:
             req.url += req.env['REPO_NAME'] + '/'
 
         if 'PATH_INFO' in req.env:
             parts = req.env['PATH_INFO'].strip('/').split('/')
             repo_parts = req.env.get('REPO_NAME', '').split('/')
             if parts[:len(repo_parts)] == repo_parts:
                 parts = parts[len(repo_parts):]
             query = '/'.join(parts)
         else:
             query = req.env['QUERY_STRING'].split('&', 1)[0]
             query = query.split(';', 1)[0]
 
         # process this if it's a protocol request
         # protocol bits don't need to create any URLs
         # and the clients always use the old URL structure
 
         cmd = req.form.get('cmd', [''])[0]
         if protocol.iscmd(cmd):
             try:
                 if query:
                     raise ErrorResponse(HTTP_NOT_FOUND)
                 if cmd in perms:
                     self.check_perm(rctx, req, perms[cmd])
                 return protocol.call(rctx.repo, req, cmd)
             except ErrorResponse as inst:
                 # A client that sends unbundle without 100-continue will
                 # break if we respond early.
                 if (cmd == 'unbundle' and
                     (req.env.get('HTTP_EXPECT',
                                  '').lower() != '100-continue') or
                     req.env.get('X-HgHttp2', '')):
                     req.drain()
                 else:
                     req.headers.append(('Connection', 'Close'))
                 req.respond(inst, protocol.HGTYPE,
                             body='0\n%s\n' % inst)
                 return ''
 
         # translate user-visible url structure to internal structure
 
         args = query.split('/', 2)
         if 'cmd' not in req.form and args and args[0]:
 
             cmd = args.pop(0)
             style = cmd.rfind('-')
             if style != -1:
                 req.form['style'] = [cmd[:style]]
                 cmd = cmd[style + 1:]
 
             # avoid accepting e.g. style parameter as command
             if util.safehasattr(webcommands, cmd):
                 req.form['cmd'] = [cmd]
 
             if cmd == 'static':
                 req.form['file'] = ['/'.join(args)]
             else:
                 if args and args[0]:
                     node = args.pop(0).replace('%2F', '/')
                     req.form['node'] = [node]
                 if args:
                     req.form['file'] = args
 
             ua = req.env.get('HTTP_USER_AGENT', '')
             if cmd == 'rev' and 'mercurial' in ua:
                 req.form['style'] = ['raw']
 
             if cmd == 'archive':
                 fn = req.form['node'][0]
                 for type_, spec in rctx.archivespecs.iteritems():
                     ext = spec[2]
                     if fn.endswith(ext):
                         req.form['node'] = [fn[:-len(ext)]]
                         req.form['type'] = [type_]
 
         # process the web interface request
 
         try:
             tmpl = rctx.templater(req)
             ctype = tmpl('mimetype', encoding=encoding.encoding)
             ctype = templater.stringify(ctype)
 
             # check read permissions non-static content
             if cmd != 'static':
                 self.check_perm(rctx, req, None)
 
             if cmd == '':
                 req.form['cmd'] = [tmpl.cache['default']]
                 cmd = req.form['cmd'][0]
 
             if rctx.configbool('web', 'cache', True):
                 caching(self, req) # sets ETag header or raises NOT_MODIFIED
             if cmd not in webcommands.__all__:
                 msg = 'no such method: %s' % cmd
                 raise ErrorResponse(HTTP_BAD_REQUEST, msg)
             elif cmd == 'file' and 'raw' in req.form.get('style', []):
                 rctx.ctype = ctype
                 content = webcommands.rawfile(rctx, req, tmpl)
             else:
                 content = getattr(webcommands, cmd)(rctx, req, tmpl)
                 req.respond(HTTP_OK, ctype)
 
             return content
 
         except (error.LookupError, error.RepoLookupError) as err:
             req.respond(HTTP_NOT_FOUND, ctype)
             msg = str(err)
             if (util.safehasattr(err, 'name') and
                 not isinstance(err,  error.ManifestLookupError)):
                 msg = 'revision not found: %s' % err.name
             return tmpl('error', error=msg)
         except (error.RepoError, error.RevlogError) as inst:
             req.respond(HTTP_SERVER_ERROR, ctype)
             return tmpl('error', error=str(inst))
         except ErrorResponse as inst:
             req.respond(inst, ctype)
             if inst.code == HTTP_NOT_MODIFIED:
                 # Not allowed to return a body on a 304
                 return ['']
             return tmpl('error', error=str(inst))
 
     def check_perm(self, rctx, req, op):
         for permhook in permhooks:
             permhook(rctx, req, op)
 
 def getwebview(repo):
     """The 'web.view' config controls changeset filter to hgweb. Possible
     values are ``served``, ``visible`` and ``all``. Default is ``served``.
     The ``served`` filter only shows changesets that can be pulled from the
     hgweb instance.  The``visible`` filter includes secret changesets but
     still excludes "hidden" one.
 
     See the repoview module for details.
 
     The option has been around undocumented since Mercurial 2.5, but no
     user ever asked about it. So we better keep it undocumented for now."""
     viewconfig = repo.ui.config('web', 'view', 'served',
                                 untrusted=True)
     if viewconfig == 'all':
         return repo.unfiltered()
     elif viewconfig in repoview.filtertable:
         return repo.filtered(viewconfig)
     else:
         return repo.filtered('served')
 

tests/test-hgweb-non-interactive.t

@@ -1,82 +1,83
 Tests if hgweb can run without touching sys.stdin, as is required
 by the WSGI standard and strictly implemented by mod_wsgi.
 
   $ hg init repo
   $ cd repo
   $ echo foo > bar
   $ hg add bar
   $ hg commit -m "test"
   $ cat > request.py <<EOF
   > from mercurial import dispatch
   > from mercurial.hgweb.hgweb_mod import hgweb
   > from mercurial.ui import ui
   > from mercurial import hg
   > from StringIO import StringIO
   > import os, sys
   > 
   > class FileLike(object):
   >     def __init__(self, real):
   >         self.real = real
   >     def fileno(self):
   >         print >> sys.__stdout__, 'FILENO'
   >         return self.real.fileno()
   >     def read(self):
   >         print >> sys.__stdout__, 'READ'
   >         return self.real.read()
   >     def readline(self):
   >         print >> sys.__stdout__, 'READLINE'
   >         return self.real.readline()
   > 
   > sys.stdin = FileLike(sys.stdin)
   > errors = StringIO()
   > input = StringIO()
   > output = StringIO()
   > 
   > def startrsp(status, headers):
   >     print '---- STATUS'
   >     print status
   >     print '---- HEADERS'
   >     print [i for i in headers if i[0] != 'ETag']
   >     print '---- DATA'
   >     return output.write
   > 
   > env = {
   >     'wsgi.version': (1, 0),
   >     'wsgi.url_scheme': 'http',
   >     'wsgi.errors': errors,
   >     'wsgi.input': input,
   >     'wsgi.multithread': False,
   >     'wsgi.multiprocess': False,
   >     'wsgi.run_once': False,
   >     'REQUEST_METHOD': 'GET',
   >     'SCRIPT_NAME': '',
   >     'PATH_INFO': '',
   >     'QUERY_STRING': '',
   >     'SERVER_NAME': '127.0.0.1',
   >     'SERVER_PORT': os.environ['HGPORT'],
   >     'SERVER_PROTOCOL': 'HTTP/1.0'
   > }
   > 
   > i = hgweb('.')
   > i(env, startrsp)
   > print '---- ERRORS'
   > print errors.getvalue()
   > print '---- OS.ENVIRON wsgi variables'
   > print sorted([x for x in os.environ if x.startswith('wsgi')])
   > print '---- request.ENVIRON wsgi variables'
-  > print sorted([x for x in i._getrepo().ui.environ if x.startswith('wsgi')])
+  > with i._obtainrepo() as repo:
+  >     print sorted([x for x in repo.ui.environ if x.startswith('wsgi')])
   > EOF
   $ python request.py
   ---- STATUS
   200 Script output follows
   ---- HEADERS
   [('Content-Type', 'text/html; charset=ascii')]
   ---- DATA
   ---- ERRORS
   
   ---- OS.ENVIRON wsgi variables
   []
   ---- request.ENVIRON wsgi variables
   ['wsgi.errors', 'wsgi.input', 'wsgi.multiprocess', 'wsgi.multithread', 'wsgi.run_once', 'wsgi.url_scheme', 'wsgi.version']
 
   $ cd ..