def test_msg_hash(self):
"""Test that ForwardMsg hash generation works as expected"""
widget_idA = "A0123456789"
widget_idB = "B0123456789"
file_name = "example_file.png"
file_bytes = bytearray(
"0123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789",
"utf-8",
)
uploaded_file_mgr = UploadedFileManager()
uploaded_file_mgr.create_or_clear_file(widget_idA, file_name,
len(file_bytes), date.today(),
1)
uploaded_file_mgr.create_or_clear_file(widget_idB, file_name,
len(file_bytes), date.today(),
2)
progress_a = uploaded_file_mgr.process_chunk(widget_idA, 0, file_bytes)
self.assertEqual(progress_a, 1)
progress_b = uploaded_file_mgr.process_chunk(widget_idB, 0,
file_bytes[0:50])
self.assertEqual(progress_b, 0.5)
progress_b = uploaded_file_mgr.process_chunk(widget_idB, 1,
file_bytes[50:100])
self.assertEqual(progress_b, 1)
progress_a, data_a = uploaded_file_mgr.get_data(widget_idA)
progress_b, data_b = uploaded_file_mgr.get_data(widget_idB)
self.assertEqual(progress_a, 100)
self.assertEqual(progress_b, 100)
self.assertEqual(len(data_a), len(file_bytes))
self.assertEqual(data_a, file_bytes)
self.assertEqual(data_a, data_b)
uploaded_file_mgr.delete_file(widget_idA)
progress_a, data_a = uploaded_file_mgr.get_data(widget_idA)
self.assertEqual(progress_a, 0)
self.assertEqual(data_a, None)
uploaded_file_mgr.delete_all_files()
progress_b, data_b = uploaded_file_mgr.get_data(widget_idB)
self.assertEqual(progress_b, 0)
self.assertEqual(data_b, None)
class ReportSession(object):
"""
Contains session data for a single "user" of an active report
(that is, a connected browser tab).
Each ReportSession has its own Report, root DeltaGenerator, ScriptRunner,
and widget state.
A ReportSession is attached to each thread involved in running its Report.
"""
_next_id = 0
def __init__(self, ioloop, script_path, command_line):
"""Initialize the ReportSession.
Parameters
----------
ioloop : tornado.ioloop.IOLoop
The Tornado IOLoop that we're running within.
script_path : str
Path of the Python file from which this report is generated.
command_line : str
Command line as input by the user.
"""
# Each ReportSession gets a unique ID
self.id = ReportSession._next_id
ReportSession._next_id += 1
self._ioloop = ioloop
self._report = Report(script_path, command_line)
self._state = ReportSessionState.REPORT_NOT_RUNNING
self._uploaded_file_mgr = UploadedFileManager()
self._widget_states = WidgetStates()
self._local_sources_watcher = LocalSourcesWatcher(
self._report, self._on_source_file_changed)
self._sent_initialize_message = False
self._storage = None
self._maybe_reuse_previous_run = False
self._run_on_save = config.get_option("server.runOnSave")
# The ScriptRequestQueue is the means by which we communicate
# with the active ScriptRunner.
self._script_request_queue = ScriptRequestQueue()
self._scriptrunner = None
LOGGER.debug("ReportSession initialized (id=%s)", self.id)
def flush_browser_queue(self):
"""Clears the report queue and returns the messages it contained.
The Server calls this periodically to deliver new messages
to the browser connected to this report.
Returns
-------
list[ForwardMsg]
The messages that were removed from the queue and should
be delivered to the browser.
"""
return self._report.flush_browser_queue()
def shutdown(self):
"""Shuts down the ReportSession.
It's an error to use a ReportSession after it's been shut down.
"""
if self._state != ReportSessionState.SHUTDOWN_REQUESTED:
LOGGER.debug("Shutting down (id=%s)", self.id)
self._uploaded_file_mgr.delete_all_files()
# Shut down the ScriptRunner, if one is active.
# self._state must not be set to SHUTDOWN_REQUESTED until
# after this is called.
if self._scriptrunner is not None:
self._enqueue_script_request(ScriptRequest.SHUTDOWN)
self._state = ReportSessionState.SHUTDOWN_REQUESTED
self._local_sources_watcher.close()
def enqueue(self, msg):
"""Enqueues a new ForwardMsg to our browser queue.
This can be called on both the main thread and a ScriptRunner
run thread.
Parameters
----------
msg : ForwardMsg
The message to enqueue
"""
if not config.get_option("client.displayEnabled"):
return
# Avoid having two maybe_handle_execution_control_request running on
# top of each other when tracer is installed. This leads to a lock
# contention.
if not config.get_option("runner.installTracer"):
# If we have an active ScriptRunner, signal that it can handle an
# execution control request. (Copy the scriptrunner reference to
# avoid it being unset from underneath us, as this function can be
# called outside the main thread.)
scriptrunner = self._scriptrunner
if scriptrunner is not None:
scriptrunner.maybe_handle_execution_control_request()
self._report.enqueue(msg)
def enqueue_exception(self, e):
"""Enqueues an Exception message.
Parameters
----------
e : BaseException
"""
# This does a few things:
# 1) Clears the current report in the browser.
# 2) Marks the current report as "stopped" in the browser.
# 3) HACK: Resets any script params that may have been broken (e.g. the
# command-line when rerunning with wrong argv[0])
self._on_scriptrunner_event(
ScriptRunnerEvent.SCRIPT_STOPPED_WITH_SUCCESS)
self._on_scriptrunner_event(ScriptRunnerEvent.SCRIPT_STARTED)
self._on_scriptrunner_event(
ScriptRunnerEvent.SCRIPT_STOPPED_WITH_SUCCESS)
msg = ForwardMsg()
msg.metadata.delta_id = 0
exception_proto.marshall(msg.delta.new_element.exception, e)
self.enqueue(msg)
def request_rerun(self, widget_state=None):
"""Signal that we're interested in running the script.
If the script is not already running, it will be started immediately.
Otherwise, a rerun will be requested.
Parameters
----------
widget_state : dict | None
The widget state dictionary to run the script with, or None
to use the widget state from the previous run of the script.
"""
self._enqueue_script_request(ScriptRequest.RERUN,
RerunData(widget_state))
def _on_source_file_changed(self):
"""One of our source files changed. Schedule a rerun if appropriate."""
if self._run_on_save:
self.request_rerun()
else:
self._enqueue_file_change_message()
def _clear_queue(self):
self._report.clear()
def _on_scriptrunner_event(self,
event,
exception=None,
widget_states=None):
"""Called when our ScriptRunner emits an event.
This is *not* called on the main thread.
Parameters
----------
event : ScriptRunnerEvent
exception : BaseException | None
An exception thrown during compilation. Set only for the
SCRIPT_STOPPED_WITH_COMPILE_ERROR event.
widget_states : streamlit.proto.Widget_pb2.WidgetStates | None
The ScriptRunner's final WidgetStates. Set only for the
SHUTDOWN event.
"""
LOGGER.debug("OnScriptRunnerEvent: %s", event)
prev_state = self._state
if event == ScriptRunnerEvent.SCRIPT_STARTED:
if self._state != ReportSessionState.SHUTDOWN_REQUESTED:
self._state = ReportSessionState.REPORT_IS_RUNNING
if config.get_option("server.liveSave"):
# Enqueue into the IOLoop so it runs without blocking AND runs
# on the main thread.
self._ioloop.spawn_callback(self._save_running_report)
self._clear_queue()
self._maybe_enqueue_initialize_message()
self._enqueue_new_report_message()
elif (event == ScriptRunnerEvent.SCRIPT_STOPPED_WITH_SUCCESS
or event == ScriptRunnerEvent.SCRIPT_STOPPED_WITH_COMPILE_ERROR):
if self._state != ReportSessionState.SHUTDOWN_REQUESTED:
self._state = ReportSessionState.REPORT_NOT_RUNNING
script_succeeded = event == ScriptRunnerEvent.SCRIPT_STOPPED_WITH_SUCCESS
self._enqueue_report_finished_message(
ForwardMsg.FINISHED_SUCCESSFULLY if script_succeeded else
ForwardMsg.FINISHED_WITH_COMPILE_ERROR)
if config.get_option("server.liveSave"):
# Enqueue into the IOLoop so it runs without blocking AND runs
# on the main thread.
self._ioloop.spawn_callback(self._save_final_report_and_quit)
if script_succeeded:
# When a script completes successfully, we update our
# LocalSourcesWatcher to account for any source code changes
# that change which modules should be watched. (This is run on
# the main thread, because LocalSourcesWatcher is not
# thread safe.)
self._ioloop.spawn_callback(
self._local_sources_watcher.update_watched_modules)
else:
# When a script fails to compile, we send along the exception.
from streamlit.elements import exception_proto
msg = ForwardMsg()
exception_proto.marshall(
msg.session_event.script_compilation_exception, exception)
self.enqueue(msg)
elif event == ScriptRunnerEvent.SHUTDOWN:
# When ScriptRunner shuts down, update our local reference to it,
# and check to see if we need to spawn a new one. (This is run on
# the main thread.)
def on_shutdown():
self._widget_states = widget_states
self._scriptrunner = None
# Because a new ScriptEvent could have been enqueued while the
# scriptrunner was shutting down, we check to see if we should
# create a new one. (Otherwise, a newly-enqueued ScriptEvent
# won't be processed until another event is enqueued.)
self._maybe_create_scriptrunner()
self._ioloop.spawn_callback(on_shutdown)
# Send a message if our run state changed
report_was_running = prev_state == ReportSessionState.REPORT_IS_RUNNING
report_is_running = self._state == ReportSessionState.REPORT_IS_RUNNING
if report_is_running != report_was_running:
self._enqueue_session_state_changed_message()
def _enqueue_session_state_changed_message(self):
msg = ForwardMsg()
msg.session_state_changed.run_on_save = self._run_on_save
msg.session_state_changed.report_is_running = (
self._state == ReportSessionState.REPORT_IS_RUNNING)
self.enqueue(msg)
def _enqueue_file_change_message(self):
LOGGER.debug("Enqueuing report_changed message (id=%s)", self.id)
msg = ForwardMsg()
msg.session_event.report_changed_on_disk = True
self.enqueue(msg)
def _maybe_enqueue_initialize_message(self):
if self._sent_initialize_message:
return
self._sent_initialize_message = True
msg = ForwardMsg()
imsg = msg.initialize
imsg.config.sharing_enabled = config.get_option(
"global.sharingMode") != "off"
imsg.config.gather_usage_stats = config.get_option(
"browser.gatherUsageStats")
imsg.config.max_cached_message_age = config.get_option(
"global.maxCachedMessageAge")
imsg.config.mapbox_token = config.get_option("mapbox.token")
LOGGER.debug(
"New browser connection: "
"gather_usage_stats=%s, "
"sharing_enabled=%s, "
"max_cached_message_age=%s",
imsg.config.gather_usage_stats,
imsg.config.sharing_enabled,
imsg.config.max_cached_message_age,
)
imsg.environment_info.streamlit_version = __version__
imsg.environment_info.python_version = ".".join(
map(str, sys.version_info))
imsg.session_state.run_on_save = self._run_on_save
imsg.session_state.report_is_running = (
self._state == ReportSessionState.REPORT_IS_RUNNING)
imsg.user_info.installation_id = __installation_id__
if Credentials.get_current().activation:
imsg.user_info.email = Credentials.get_current().activation.email
else:
imsg.user_info.email = ""
imsg.command_line = self._report.command_line
self.enqueue(msg)
def _enqueue_new_report_message(self):
self._report.generate_new_id()
msg = ForwardMsg()
msg.new_report.id = self._report.report_id
msg.new_report.name = self._report.name
msg.new_report.script_path = self._report.script_path
self.enqueue(msg)
def _enqueue_report_finished_message(self, status):
"""Enqueues a report_finished ForwardMsg.
Parameters
----------
status : ReportFinishedStatus
"""
msg = ForwardMsg()
msg.report_finished = status
self.enqueue(msg)
def handle_rerun_script_request(self,
command_line=None,
widget_state=None,
is_preheat=False):
"""Tells the ScriptRunner to re-run its report.
Parameters
----------
command_line : str | None
The new command line arguments to run the script with, or None
to use its previous command line value.
widget_state : WidgetStates | None
The WidgetStates protobuf to run the script with, or None
to use its previous widget states.
is_preheat: boolean
True if this ReportSession should run the script immediately, and
then ignore the next rerun request if it matches the already-ran
widget state.
"""
if is_preheat:
self._maybe_reuse_previous_run = True # For next time.
elif self._maybe_reuse_previous_run:
# If this is a "preheated" ReportSession, reuse the previous run if
# the widget state matches. But only do this one time ever.
self._maybe_reuse_previous_run = False
has_widget_state = (widget_state is not None
and len(widget_state.widgets) > 0)
if not has_widget_state:
LOGGER.debug(
"Skipping rerun since the preheated run is the same")
return
self.request_rerun(widget_state)
def handle_upload_file(self, upload_file):
self._uploaded_file_mgr.create_or_clear_file(
widget_id=upload_file.widget_id,
name=upload_file.name,
size=upload_file.size,
last_modified=upload_file.lastModified,
chunks=upload_file.chunks,
)
self.handle_rerun_script_request(widget_state=self._widget_states)
def handle_upload_file_chunk(self, upload_file_chunk):
progress = self._uploaded_file_mgr.process_chunk(
widget_id=upload_file_chunk.widget_id,
index=upload_file_chunk.index,
data=upload_file_chunk.data,
)
if progress == 1:
self.handle_rerun_script_request(widget_state=self._widget_states)
def handle_delete_uploaded_file(self, delete_uploaded_file):
self._uploaded_file_mgr.delete_file(
widget_id=delete_uploaded_file.widget_id)
self.handle_rerun_script_request(widget_state=self._widget_states)
def handle_stop_script_request(self):
"""Tells the ScriptRunner to stop running its report."""
self._enqueue_script_request(ScriptRequest.STOP)
def handle_clear_cache_request(self):
"""Clears this report's cache.
Because this cache is global, it will be cleared for all users.
"""
# Setting verbose=True causes clear_cache to print to stdout.
# Since this command was initiated from the browser, the user
# doesn't need to see the results of the command in their
# terminal.
caching.clear_cache()
def handle_set_run_on_save_request(self, new_value):
"""Changes our run_on_save flag to the given value.
The browser will be notified of the change.
Parameters
----------
new_value : bool
New run_on_save value
"""
self._run_on_save = new_value
self._enqueue_session_state_changed_message()
def _enqueue_script_request(self, request, data=None):
"""Enqueue a ScriptEvent into our ScriptEventQueue.
If a script thread is not already running, one will be created
to handle the event.
Parameters
----------
request : ScriptRequest
The type of request.
data : Any
Data associated with the request, if any.
"""
if self._state == ReportSessionState.SHUTDOWN_REQUESTED:
LOGGER.warning("Discarding %s request after shutdown" % request)
return
self._script_request_queue.enqueue(request, data)
self._maybe_create_scriptrunner()
def _maybe_create_scriptrunner(self):
"""Create a new ScriptRunner if we have unprocessed script requests.
This is called every time a ScriptRequest is enqueued, and also
after a ScriptRunner shuts down, in case new requests were enqueued
during its termination.
This function should only be called on the main thread.
"""
if (self._state == ReportSessionState.SHUTDOWN_REQUESTED
or self._scriptrunner is not None
or not self._script_request_queue.has_request):
return
# Create the ScriptRunner, attach event handlers, and start it
self._scriptrunner = ScriptRunner(
report=self._report,
enqueue_forward_msg=self.enqueue,
widget_states=self._widget_states,
request_queue=self._script_request_queue,
uploaded_file_mgr=self._uploaded_file_mgr,
)
self._scriptrunner.on_event.connect(self._on_scriptrunner_event)
self._scriptrunner.start()
@tornado.gen.coroutine
def handle_save_request(self, ws):
"""Save serialized version of report deltas to the cloud.
"Progress" ForwardMsgs will be sent to the client during the upload.
These messages are sent "out of band" - that is, they don't get
enqueued into the ReportQueue (because they're not part of the report).
Instead, they're written directly to the report's WebSocket.
Parameters
----------
ws : _BrowserWebSocketHandler
The report's websocket handler.
"""
@tornado.gen.coroutine
def progress(percent):
progress_msg = ForwardMsg()
progress_msg.upload_report_progress = percent
yield ws.write_message(serialize_forward_msg(progress_msg),
binary=True)
# Indicate that the save is starting.
try:
yield progress(0)
url = yield self._save_final_report(progress)
# Indicate that the save is done.
progress_msg = ForwardMsg()
progress_msg.report_uploaded = url
yield ws.write_message(serialize_forward_msg(progress_msg),
binary=True)
except Exception as e:
# Horrible hack to show something if something breaks.
err_msg = "%s: %s" % (type(e).__name__, str(e)
or "No further details.")
progress_msg = ForwardMsg()
progress_msg.report_uploaded = err_msg
yield ws.write_message(serialize_forward_msg(progress_msg),
binary=True)
LOGGER.warning("Failed to save report:", exc_info=e)
@tornado.gen.coroutine
def _save_running_report(self):
files = self._report.serialize_running_report_to_files()
url = yield self._get_storage().save_report_files(
self._report.report_id, files)
if config.get_option("server.liveSave"):
url_util.print_url("Saved running app", url)
raise tornado.gen.Return(url)
@tornado.gen.coroutine
def _save_final_report(self, progress_coroutine=None):
files = self._report.serialize_final_report_to_files()
url = yield self._get_storage().save_report_files(
self._report.report_id, files, progress_coroutine)
if config.get_option("server.liveSave"):
url_util.print_url("Saved final app", url)
raise tornado.gen.Return(url)
@tornado.gen.coroutine
def _save_final_report_and_quit(self):
yield self._save_final_report()
self._ioloop.stop()
def _get_storage(self):
if self._storage is None:
sharing_mode = config.get_option("global.sharingMode")
if sharing_mode == "s3":
self._storage = S3Storage()
elif sharing_mode == "file":
self._storage = FileStorage()
else:
raise RuntimeError("Unsupported sharing mode '%s'" %
sharing_mode)
return self._storage