optParameters = [
["node-directory", "d", None, "Specify the directory of the node to be run. [default, for 'tahoe run' only: current directory]"],
- ["multiple", "m", None, "['tahoe run' cannot accept multiple node directories]"],
- ]
-
-def do_start(basedir, opts, out=sys.stdout, err=sys.stderr):
+ ]
+ optFlags = [ ]
+ if BasedirMixin.can_start_multiple:
+ # usage.Options doesn't let us remove flags that we inherit from a
+ # parent class, so at least provide a --help string that warns people
+ # away from using it. There is also code (switching on
+ # allow_multiple) to disable it at runtime.
+ optFlags.append(["multiple", "m",
+ "['tahoe run' cannot accept multiple node directories]"])
+
+def do_start(basedir, opts, out=sys.stdout, err=sys.stderr, fork=False):
print >>out, "STARTING", quote_output(basedir)
if not os.path.isdir(basedir):
print >>err, "%s does not look like a directory at all" % quote_output(basedir)
if opts["profile"]:
args.extend(["--profile=profiling_results.prof", "--savestats",])
# now we're committed
+ if fork:
+ if os.fork() != 0:
+ return 0 # parent
+ # we're in the child
os.chdir(basedir)
from twisted.scripts import twistd
sys.argv = args
def start(config, stdout, stderr):
rc = 0
- for basedir in config['basedirs']:
- rc = do_start(basedir, config, stdout, stderr) or rc
+ for basedir in config['basedirs'][:-1]:
+ # fork before starting all but the last one
+ rc = do_start(basedir, config, stdout, stderr, fork=True) or rc
+ # start the last one in the current process, to capture its exit code
+ rc = do_start(config['basedirs'][-1], config, stdout, stderr, fork=False) or rc
return rc
def stop(config, stdout, stderr):
if rc:
print >>stderr, "not restarting"
return rc
- for basedir in config['basedirs']:
- rc = do_start(basedir, config, stdout, stderr) or rc
+ rc = start(config, stdout, stderr) or rc
return rc
def run(config, stdout, stderr):
d.addBoth(_remove_hotline)
return d
+ def test_multiple_clients(self):
+ self.skip_if_cannot_daemonize()
+ basedir = self.workdir("test_multiple_clients")
+ # this furl must be syntactically correct, else 'tahoe start' will
+ # correctly report an error. It doesn't need to point at a real
+ # introducer.
+ introducer_furl = "pb://xrndsskn2zuuian5ltnxrte7lnuqdrkz@127.0.0.1:55617/introducer"
+ c1 = os.path.join(basedir, "c1")
+ HOTLINE_FILE_1 = os.path.join(c1, "suicide_prevention_hotline")
+ TWISTD_PID_FILE_1 = os.path.join(c1, "twistd.pid")
+ PORTNUMFILE_1 = os.path.join(c1, "client.port")
+ c1_args = ["--quiet", "create-node", "--basedir", c1, "--webport", "0",
+ "--introducer", introducer_furl]
+
+ c2 = os.path.join(basedir, "c2")
+ HOTLINE_FILE_2 = os.path.join(c2, "suicide_prevention_hotline")
+ TWISTD_PID_FILE_2 = os.path.join(c2, "twistd.pid")
+ PORTNUMFILE_2 = os.path.join(c2, "client.port")
+ c2_args = ["--quiet", "create-node", "--basedir", c2, "--webport", "0",
+ "--introducer", introducer_furl]
+
+ d = utils.getProcessOutputAndValue(bintahoe, args=c1_args, env=os.environ)
+ def _cb1(res):
+ out, err, rc_or_sig = res
+ self.failUnlessEqual(rc_or_sig, 0)
+ return utils.getProcessOutputAndValue(bintahoe, args=c2_args,
+ env=os.environ)
+ d.addCallback(_cb1)
+ def _cb2(res):
+ out, err, rc_or_sig = res
+ self.failUnlessEqual(rc_or_sig, 0)
+ # both nodes have been constructed, but they are not yet running
+
+ # By writing this file, we get sixty seconds before the client
+ # will exit. This insures that even if the 'stop' command doesn't
+ # work (and the test fails), the client should still terminate.
+ open(HOTLINE_FILE_1, "w").write("")
+ open(HOTLINE_FILE_2, "w").write("")
+ # now it's safe to start the nodes
+ start_args = ["--quiet", "start", "-m", c1, c2]
+ return utils.getProcessOutputAndValue(bintahoe, args=start_args,
+ env=os.environ)
+ d.addCallback(_cb2)
+
+ def _nodes_have_started():
+ open(HOTLINE_FILE_1, "w").write("")
+ open(HOTLINE_FILE_2, "w").write("")
+ return (os.path.exists(PORTNUMFILE_1)
+ and os.path.exists(PORTNUMFILE_2))
+
+ def _cb3(res):
+ out, err, rc_or_sig = res
+ open(HOTLINE_FILE_1, "w").write("")
+ open(HOTLINE_FILE_2, "w").write("")
+ errstr = "rc=%d, OUT: '%s', ERR: '%s'" % (rc_or_sig, out, err)
+ self.failUnlessEqual(rc_or_sig, 0, errstr)
+ self.failUnlessEqual(out, "", errstr)
+ # See test_client_no_noise -- for now we ignore noise.
+ # self.failUnlessEqual(err, "", errstr)
+
+ # if 'tahoe start' exited without error, we're guaranteed that it
+ # successfully loaded the code and tahoe.cfg for the last node
+ # (argv[-1]), since these happen before twistd.run() calls
+ # fork(). There made have been a runtime error after fork().
+ # Also, there may have been an import or config error while
+ # loading any of the earlier nodes (argv[:-1]), since 'tahoe
+ # start' must do an additional fork before calling twistd.run()
+ # on those.
+
+ # so watch the filesystem to determine when the nodes have
+ # started running
+
+ return self.poll(_nodes_have_started)
+ d.addCallback(_cb3)
+
+ def _restart(ign):
+ # delete the port-number files, then restart the nodes. The new
+ # instances will re-write those files.
+ os.unlink(PORTNUMFILE_1)
+ os.unlink(PORTNUMFILE_2)
+ restart_args = ["--quiet", "restart", "-m", c1, c2]
+ return utils.getProcessOutputAndValue(bintahoe, args=restart_args,
+ env=os.environ)
+ d.addCallback(_restart)
+ def _did_restart(res):
+ out, err, rc_or_sig = res
+ errstr = "rc=%d, OUT: '%s', ERR: '%s'" % (rc_or_sig, out, err)
+ self.failUnlessEqual(rc_or_sig, 0, errstr)
+ self.failUnlessEqual(out, "", errstr)
+ # See test_client_no_noise -- for now we ignore noise.
+ # self.failUnlessEqual(err, "", errstr)
+
+ # we must still poll for the startup process to get far enough
+ return self.poll(_nodes_have_started)
+ d.addCallback(_did_restart)
+ # when that finishes, we know the nodes have restarted. If 'tahoe
+ # restart -m' is broken, that will time out.
+
+ # now we can kill it. TODO: On a slow machine, the node might kill
+ # itself before we get a chance too, especially if spawning the
+ # 'tahoe stop' command takes a while.
+ def _stop(res):
+ open(HOTLINE_FILE_1, "w").write("")
+ open(HOTLINE_FILE_2, "w").write("")
+ self.failUnless(os.path.exists(TWISTD_PID_FILE_1),
+ (TWISTD_PID_FILE_1, os.listdir(c1)))
+ self.failUnless(os.path.exists(TWISTD_PID_FILE_2),
+ (TWISTD_PID_FILE_2, os.listdir(c2)))
+ stop_args = ["--quiet", "stop", "-m", c1, c2]
+ return utils.getProcessOutputAndValue(bintahoe, args=stop_args,
+ env=os.environ)
+ d.addCallback(_stop)
+
+ def _cb4(res):
+ out, err, rc_or_sig = res
+ errstr = "rc=%d, OUT: '%s', ERR: '%s'" % (rc_or_sig, out, err)
+ self.failUnlessEqual(rc_or_sig, 0, errstr)
+ self.failUnlessEqual(out, "", errstr)
+ # See test_client_no_noise -- for now we ignore noise.
+ # self.failUnlessEqual(err, "", errstr)
+
+ # the parent was supposed to poll and wait until it sees
+ # twistd.pid go away before it exits, so twistd.pid should be
+ # gone by now.
+ self.failIf(os.path.exists(TWISTD_PID_FILE_1))
+ self.failIf(os.path.exists(TWISTD_PID_FILE_2))
+ d.addCallback(_cb4)
+ def _remove_hotline(res):
+ # always remove these, so that if something went wrong, the nodes
+ # will quit eventually
+ os.unlink(HOTLINE_FILE_1)
+ os.unlink(HOTLINE_FILE_2)
+ return res
+ d.addBoth(_remove_hotline)
+ return d
+
+
def test_baddir(self):
self.skip_if_cannot_daemonize()
basedir = self.workdir("test_baddir")