def backup_unprocessed_event(project, data):
"""
Backup unprocessed event payload into redis. Only call if event should be
able to be reprocessed.
"""
if not features.has("projects:reprocessing-v2", project, actor=None):
return
event_processing_store.store(data, unprocessed=True)
def backup_unprocessed_event(project, data):
"""
Backup unprocessed event payload into redis. Only call if event should be
able to be reprocessed.
"""
if options.get("store.reprocessing-force-disable"):
return
event_processing_store.store(dict(data), unprocessed=True)
def backup_unprocessed_event(project, data):
"""
Backup unprocessed event payload into redis. Only call if event should be
able to be reprocessed.
"""
if features.has("organizations:reprocessing-v2",
project.organization,
actor=None) or _should_capture_nodestore_stats(
data.get("event_id")):
event_processing_store.store(data, unprocessed=True)
def inner(data):
data.setdefault("timestamp", iso_format(before_now(seconds=1)))
mgr = EventManager(data=data, project=default_project)
mgr.normalize()
data = mgr.get_data()
event_id = data["event_id"]
cache_key = event_processing_store.store(data)
with task_runner():
# factories.store_event would almost be suitable for this, but let's
# actually run through stacktrace processing once
preprocess_event(start_time=time(), cache_key=cache_key, data=data)
return event_id
def inner(data, seconds_ago=1):
# Set platform to native so all parts of reprocessing fire, symbolication will
# not happen without this set to certain values
data.setdefault("platform", "native")
# Every request to snuba has a timestamp that's clamped in a curious way to
# ensure data consistency
data.setdefault("timestamp",
iso_format(before_now(seconds=seconds_ago)))
mgr = EventManager(data=data, project=default_project)
mgr.normalize()
data = mgr.get_data()
event_id = data["event_id"]
cache_key = event_processing_store.store(data)
with task_runner():
# factories.store_event would almost be suitable for this, but let's
# actually run through stacktrace processing once
preprocess_event(start_time=time(), cache_key=cache_key, data=data)
return event_id
def _do_process_event(
cache_key,
start_time,
event_id,
process_task,
data=None,
data_has_changed=None,
from_symbolicate=False,
):
from sentry.plugins.base import plugins
if data is None:
data = event_processing_store.get(cache_key)
if data is None:
metrics.incr("events.failed",
tags={
"reason": "cache",
"stage": "process"
},
skip_internal=False)
error_logger.error("process.failed.empty",
extra={"cache_key": cache_key})
return
data = CanonicalKeyDict(data)
project_id = data["project"]
set_current_event_project(project_id)
event_id = data["event_id"]
if killswitch_matches_context(
"store.load-shed-process-event-projects",
{
"project_id": project_id,
"event_id": event_id,
"platform": data.get("platform") or "null",
},
):
return
with sentry_sdk.start_span(
op="tasks.store.process_event.get_project_from_cache"):
project = Project.objects.get_from_cache(id=project_id)
with metrics.timer(
"tasks.store.process_event.organization.get_from_cache"):
project._organization_cache = Organization.objects.get_from_cache(
id=project.organization_id)
has_changed = bool(data_has_changed)
with sentry_sdk.start_span(
op="tasks.store.process_event.get_reprocessing_revision"):
# Fetch the reprocessing revision
reprocessing_rev = reprocessing.get_reprocessing_revision(project_id)
# Stacktrace based event processors.
with sentry_sdk.start_span(op="task.store.process_event.stacktraces"):
with metrics.timer("tasks.store.process_event.stacktraces",
tags={"from_symbolicate": from_symbolicate}):
new_data = process_stacktraces(data)
if new_data is not None:
has_changed = True
data = new_data
# Second round of datascrubbing after stacktrace and language-specific
# processing. First round happened as part of ingest.
#
# *Right now* the only sensitive data that is added in stacktrace
# processing are usernames in filepaths, so we run directly after
# stacktrace processors.
#
# We do not yet want to deal with context data produced by plugins like
# sessionstack or fullstory (which are in `get_event_preprocessors`), as
# this data is very unlikely to be sensitive data. This is why scrubbing
# happens somewhere in the middle of the pipeline.
#
# On the other hand, Javascript event error translation is happening after
# this block because it uses `get_event_preprocessors` instead of
# `get_event_enhancers`.
#
# We are fairly confident, however, that this should run *before*
# re-normalization as it is hard to find sensitive data in partially
# trimmed strings.
if has_changed and options.get("processing.can-use-scrubbers"):
with sentry_sdk.start_span(op="task.store.datascrubbers.scrub"):
with metrics.timer("tasks.store.datascrubbers.scrub",
tags={"from_symbolicate": from_symbolicate}):
new_data = safe_execute(scrub_data,
project=project,
event=data.data)
# XXX(markus): When datascrubbing is finally "totally stable", we might want
# to drop the event if it crashes to avoid saving PII
if new_data is not None:
data.data = new_data
# TODO(dcramer): ideally we would know if data changed by default
# Default event processors.
for plugin in plugins.all(version=2):
with sentry_sdk.start_span(
op="task.store.process_event.preprocessors") as span:
span.set_data("plugin", plugin.slug)
span.set_data("from_symbolicate", from_symbolicate)
with metrics.timer(
"tasks.store.process_event.preprocessors",
tags={
"plugin": plugin.slug,
"from_symbolicate": from_symbolicate
},
):
processors = safe_execute(plugin.get_event_preprocessors,
data=data,
_with_transaction=False)
for processor in processors or ():
try:
result = processor(data)
except Exception:
error_logger.exception(
"tasks.store.preprocessors.error")
data.setdefault("_metrics",
{})["flag.processing.error"] = True
has_changed = True
else:
if result:
data = result
has_changed = True
assert data[
"project"] == project_id, "Project cannot be mutated by plugins"
# We cannot persist canonical types in the cache, so we need to
# downgrade this.
if isinstance(data, CANONICAL_TYPES):
data = dict(data.items())
if has_changed:
# Run some of normalization again such that we don't:
# - persist e.g. incredibly large stacktraces from minidumps
# - store event timestamps that are older than our retention window
# (also happening with minidumps)
normalizer = StoreNormalizer(remove_other=False,
is_renormalize=True,
**DEFAULT_STORE_NORMALIZER_ARGS)
data = normalizer.normalize_event(dict(data))
issues = data.get("processing_issues")
try:
if issues and create_failed_event(
cache_key,
data,
project_id,
list(issues.values()),
event_id=event_id,
start_time=start_time,
reprocessing_rev=reprocessing_rev,
):
return
except RetryProcessing:
# If `create_failed_event` indicates that we need to retry we
# invoke ourselves again. This happens when the reprocessing
# revision changed while we were processing.
_do_preprocess_event(cache_key, data, start_time, event_id,
process_task, project)
return
cache_key = event_processing_store.store(data)
from_reprocessing = process_task is process_event_from_reprocessing
submit_save_event(project, from_reprocessing, cache_key, event_id,
start_time, data)
def _do_symbolicate_event(cache_key,
start_time,
event_id,
symbolicate_task,
data=None):
from sentry.lang.native.processing import get_symbolication_function
if data is None:
data = event_processing_store.get(cache_key)
if data is None:
metrics.incr("events.failed",
tags={
"reason": "cache",
"stage": "symbolicate"
},
skip_internal=False)
error_logger.error("symbolicate.failed.empty",
extra={"cache_key": cache_key})
return
data = CanonicalKeyDict(data)
project_id = data["project"]
set_current_event_project(project_id)
event_id = data["event_id"]
if killswitch_matches_context(
"store.load-shed-symbolicate-event-projects",
{
"project_id": project_id,
"event_id": event_id,
"platform": data.get("platform") or "null",
},
):
return
symbolication_function = get_symbolication_function(data)
has_changed = False
from_reprocessing = symbolicate_task is symbolicate_event_from_reprocessing
symbolication_start_time = time()
with sentry_sdk.start_span(
op="tasks.store.symbolicate_event.symbolication") as span:
span.set_data("symbolicaton_function", symbolication_function.__name__)
with metrics.timer(
"tasks.store.symbolicate_event.symbolication",
tags={
"symbolication_function": symbolication_function.__name__
},
):
while True:
try:
with sentry_sdk.start_span(
op="tasks.store.symbolicate_event.%s" %
symbolication_function.__name__) as span:
symbolicated_data = symbolication_function(data)
span.set_data("symbolicated_data",
bool(symbolicated_data))
if symbolicated_data:
data = symbolicated_data
has_changed = True
break
except RetrySymbolication as e:
if (time() - symbolication_start_time
) > settings.SYMBOLICATOR_PROCESS_EVENT_WARN_TIMEOUT:
error_logger.warning(
"symbolicate.slow",
extra={
"project_id": project_id,
"event_id": event_id
},
)
if (time() - symbolication_start_time
) > settings.SYMBOLICATOR_PROCESS_EVENT_HARD_TIMEOUT:
# Do not drop event but actually continue with rest of pipeline
# (persisting unsymbolicated event)
metrics.incr(
"tasks.store.symbolicate_event.fatal",
tags={
"reason":
"timeout",
"symbolication_function":
symbolication_function.__name__,
},
)
error_logger.exception(
"symbolicate.failed.infinite_retry",
extra={
"project_id": project_id,
"event_id": event_id
},
)
data.setdefault("_metrics",
{})["flag.processing.error"] = True
data.setdefault("_metrics",
{})["flag.processing.fatal"] = True
has_changed = True
break
else:
# sleep for `retry_after` but max 5 seconds and try again
metrics.incr(
"tasks.store.symbolicate_event.retry",
tags={
"symbolication_function":
symbolication_function.__name__
},
)
sleep(min(e.retry_after, SYMBOLICATOR_MAX_RETRY_AFTER))
continue
except Exception:
metrics.incr(
"tasks.store.symbolicate_event.fatal",
tags={
"reason":
"error",
"symbolication_function":
symbolication_function.__name__,
},
)
error_logger.exception(
"tasks.store.symbolicate_event.symbolication")
data.setdefault("_metrics",
{})["flag.processing.error"] = True
data.setdefault("_metrics",
{})["flag.processing.fatal"] = True
has_changed = True
break
# We cannot persist canonical types in the cache, so we need to
# downgrade this.
if isinstance(data, CANONICAL_TYPES):
data = dict(data.items())
if has_changed:
cache_key = event_processing_store.store(data)
process_task = process_event_from_reprocessing if from_reprocessing else process_event
_do_process_event(
cache_key=cache_key,
start_time=start_time,
event_id=event_id,
process_task=process_task,
data=data,
data_has_changed=has_changed,
from_symbolicate=True,
)
def write_event_to_cache(event):
cache_data = event.data
cache_data["event_id"] = event.event_id
cache_data["project"] = event.project_id
return event_processing_store.store(cache_data)
def _do_process_event(message, projects):
payload = message["payload"]
start_time = float(message["start_time"])
event_id = message["event_id"]
project_id = int(message["project_id"])
remote_addr = message.get("remote_addr")
attachments = message.get("attachments") or ()
# check that we haven't already processed this event (a previous instance of the forwarder
# died before it could commit the event queue offset)
#
# XXX(markus): I believe this code is extremely broken:
#
# * it practically uses memcached in prod which has no consistency
# guarantees (no idea how we don't run into issues there)
#
# * a TTL of 1h basically doesn't guarantee any deduplication at all. It
# just guarantees a good error message... for one hour.
#
# This code has been ripped from the old python store endpoint. We're
# keeping it around because it does provide some protection against
# reprocessing good events if a single consumer is in a restart loop.
deduplication_key = f"ev:{project_id}:{event_id}"
if cache.get(deduplication_key) is not None:
logger.warning(
"pre-process-forwarder detected a duplicated event" " with id:%s for project:%s.",
event_id,
project_id,
)
return # message already processed do not reprocess
try:
project = projects[project_id]
except KeyError:
logger.error("Project for ingested event does not exist: %s", project_id)
return
# Parse the JSON payload. This is required to compute the cache key and
# call process_event. The payload will be put into Kafka raw, to avoid
# serializing it again.
# XXX: Do not use CanonicalKeyDict here. This may break preprocess_event
# which assumes that data passed in is a raw dictionary.
data = json.loads(payload)
cache_key = event_processing_store.store(data)
if attachments:
attachment_objects = [
CachedAttachment(type=attachment.pop("attachment_type"), **attachment)
for attachment in attachments
]
attachment_cache.set(cache_key, attachments=attachment_objects, timeout=CACHE_TIMEOUT)
# Preprocess this event, which spawns either process_event or
# save_event. Pass data explicitly to avoid fetching it again from the
# cache.
with sentry_sdk.start_span(op="ingest_consumer.process_event.preprocess_event"):
preprocess_event(
cache_key=cache_key,
data=data,
start_time=start_time,
event_id=event_id,
project=project,
)
# remember for an 1 hour that we saved this event (deduplication protection)
cache.set(deduplication_key, "", CACHE_TIMEOUT)
# emit event_accepted once everything is done
event_accepted.send_robust(ip=remote_addr, data=data, project=project, sender=process_event)
def reprocess_event(project_id, event_id, start_time):
node_id = _generate_unprocessed_event_node_id(project_id=project_id,
event_id=event_id)
with sentry_sdk.start_span(op="reprocess_events.nodestore.get"):
data = nodestore.get(node_id)
if data is None:
return
from sentry.event_manager import set_tag
from sentry.tasks.store import preprocess_event_from_reprocessing
from sentry.ingest.ingest_consumer import CACHE_TIMEOUT
# Take unprocessed data from old event and save it as unprocessed data
# under a new event ID. The second step happens in pre-process. We could
# save the "original event ID" instead and get away with writing less to
# nodestore, but doing it this way makes the logic slightly simpler.
# Step 1: Fix up the event payload for reprocessing and put it in event
# cache/event_processing_store
orig_event_id = data["event_id"]
set_tag(data, "original_event_id", orig_event_id)
event = eventstore.get_event_by_id(project_id, orig_event_id)
if event is None:
return
set_tag(data, "original_group_id", event.group_id)
# XXX: reuse event IDs
event_id = data["event_id"] = uuid.uuid4().hex
cache_key = event_processing_store.store(data)
# Step 2: Copy attachments into attachment cache
queryset = models.EventAttachment.objects.filter(
project_id=project_id, event_id=orig_event_id).select_related("file")
attachment_objects = []
for attachment_id, attachment in enumerate(queryset):
with sentry_sdk.start_span(
op="reprocess_event._copy_attachment_into_cache") as span:
span.set_data("attachment_id", attachment.id)
attachment_objects.append(
_copy_attachment_into_cache(
attachment_id=attachment_id,
attachment=attachment,
cache_key=cache_key,
cache_timeout=CACHE_TIMEOUT,
))
if attachment_objects:
with sentry_sdk.start_span(op="reprocess_event.set_attachment_meta"):
attachment_cache.set(cache_key,
attachments=attachment_objects,
timeout=CACHE_TIMEOUT)
preprocess_event_from_reprocessing(cache_key=cache_key,
start_time=start_time,
event_id=event_id)
def _store_event(data) -> str:
return event_processing_store.store(data)
def reprocess_event(project_id, event_id, start_time):
from sentry.ingest.ingest_consumer import CACHE_TIMEOUT
from sentry.lang.native.processing import get_required_attachment_types
from sentry.tasks.store import preprocess_event_from_reprocessing
with sentry_sdk.start_span(op="reprocess_events.nodestore.get"):
node_id = Event.generate_node_id(project_id, event_id)
data = nodestore.get(node_id, subkey="unprocessed")
if data is None:
node_id = _generate_unprocessed_event_node_id(project_id=project_id, event_id=event_id)
data = nodestore.get(node_id)
if data is None:
raise CannotReprocess("reprocessing_nodestore.not_found")
with sentry_sdk.start_span(op="reprocess_events.eventstore.get"):
event = eventstore.get_event_by_id(project_id, event_id)
if event is None:
raise CannotReprocess("event.not_found")
required_attachment_types = get_required_attachment_types(data)
attachments = list(
models.EventAttachment.objects.filter(
project_id=project_id, event_id=event_id, type__in=list(required_attachment_types)
)
)
missing_attachment_types = required_attachment_types - {ea.type for ea in attachments}
if missing_attachment_types:
raise CannotReprocess(
f"attachment.not_found.{'_and_'.join(sorted(missing_attachment_types))}"
)
# Step 1: Fix up the event payload for reprocessing and put it in event
# cache/event_processing_store
set_path(data, "contexts", "reprocessing", "original_issue_id", value=event.group_id)
set_path(
data, "contexts", "reprocessing", "original_primary_hash", value=event.get_primary_hash()
)
cache_key = event_processing_store.store(data)
# Step 2: Copy attachments into attachment cache. Note that we can only
# consider minidumps because filestore just stays as-is after reprocessing
# (we simply update group_id on the EventAttachment models in post_process)
attachment_objects = []
files = {f.id: f for f in models.File.objects.filter(id__in=[ea.file_id for ea in attachments])}
for attachment_id, attachment in enumerate(attachments):
with sentry_sdk.start_span(op="reprocess_event._copy_attachment_into_cache") as span:
span.set_data("attachment_id", attachment.id)
attachment_objects.append(
_copy_attachment_into_cache(
attachment_id=attachment_id,
attachment=attachment,
file=files[attachment.file_id],
cache_key=cache_key,
cache_timeout=CACHE_TIMEOUT,
)
)
if attachment_objects:
with sentry_sdk.start_span(op="reprocess_event.set_attachment_meta"):
attachment_cache.set(cache_key, attachments=attachment_objects, timeout=CACHE_TIMEOUT)
preprocess_event_from_reprocessing(
cache_key=cache_key,
start_time=start_time,
event_id=event_id,
data=data,
)
def _do_symbolicate_event(cache_key, start_time, event_id, symbolicate_task, data=None):
from sentry.lang.native.processing import get_symbolication_function
if data is None:
data = event_processing_store.get(cache_key)
if data is None:
metrics.incr(
"events.failed", tags={"reason": "cache", "stage": "symbolicate"}, skip_internal=False
)
error_logger.error("symbolicate.failed.empty", extra={"cache_key": cache_key})
return
data = CanonicalKeyDict(data)
project_id = data["project"]
set_current_project(project_id)
event_id = data["event_id"]
symbolication_function = get_symbolication_function(data)
has_changed = False
from_reprocessing = symbolicate_task is symbolicate_event_from_reprocessing
try:
with sentry_sdk.start_span(op="tasks.store.symbolicate_event.symbolication") as span:
span.set_data("symbolicaton_function", symbolication_function.__name__)
with metrics.timer("tasks.store.symbolicate_event.symbolication"):
symbolicated_data = symbolication_function(data)
span.set_data("symbolicated_data", bool(symbolicated_data))
if symbolicated_data:
data = symbolicated_data
has_changed = True
except RetrySymbolication as e:
if start_time and (time() - start_time) > settings.SYMBOLICATOR_PROCESS_EVENT_WARN_TIMEOUT:
error_logger.warning(
"symbolicate.slow", extra={"project_id": project_id, "event_id": event_id}
)
if start_time and (time() - start_time) > settings.SYMBOLICATOR_PROCESS_EVENT_HARD_TIMEOUT:
# Do not drop event but actually continue with rest of pipeline
# (persisting unsymbolicated event)
error_logger.exception(
"symbolicate.failed.infinite_retry",
extra={"project_id": project_id, "event_id": event_id},
)
data.setdefault("_metrics", {})["flag.processing.error"] = True
data.setdefault("_metrics", {})["flag.processing.fatal"] = True
has_changed = True
else:
# Requeue the task in the "sleep" queue
retry_symbolicate_event.apply_async(
args=(),
kwargs={
"symbolicate_task_name": symbolicate_task.__name__,
"task_kwargs": {
"cache_key": cache_key,
"event_id": event_id,
"start_time": start_time,
},
},
countdown=e.retry_after,
)
return
except Exception:
error_logger.exception("tasks.store.symbolicate_event.symbolication")
data.setdefault("_metrics", {})["flag.processing.error"] = True
data.setdefault("_metrics", {})["flag.processing.fatal"] = True
has_changed = True
# We cannot persist canonical types in the cache, so we need to
# downgrade this.
if isinstance(data, CANONICAL_TYPES):
data = dict(data.items())
if has_changed:
cache_key = event_processing_store.store(data)
process_task = process_event_from_reprocessing if from_reprocessing else process_event
_do_process_event(
cache_key=cache_key,
start_time=start_time,
event_id=event_id,
process_task=process_task,
data=data,
data_has_changed=has_changed,
from_symbolicate=True,
)
def get_cache_key(event):
cache_data = event.data
cache_data["event_id"] = event.event_id
cache_data["project"] = event.project_id
return event_processing_store.store(cache_data)
def _do_symbolicate_event(
cache_key, start_time, event_id, symbolicate_task, data=None, queue_switches=0
):
from sentry.lang.native.processing import get_symbolication_function
if data is None:
data = event_processing_store.get(cache_key)
if data is None:
metrics.incr(
"events.failed", tags={"reason": "cache", "stage": "symbolicate"}, skip_internal=False
)
error_logger.error("symbolicate.failed.empty", extra={"cache_key": cache_key})
return
data = CanonicalKeyDict(data)
project_id = data["project"]
set_current_event_project(project_id)
event_id = data["event_id"]
from_reprocessing = (
symbolicate_task is symbolicate_event_from_reprocessing
or symbolicate_task is symbolicate_event_from_reprocessing_low_priority
)
# check whether the event is in the wrong queue and if so, move it to the other one.
# we do this at most SYMBOLICATOR_MAX_QUEUE_SWITCHES times.
if queue_switches >= SYMBOLICATOR_MAX_QUEUE_SWITCHES:
metrics.gauge("tasks.store.symbolicate_event.low_priority.max_queue_switches", 1)
else:
is_low_priority = symbolicate_task in [
symbolicate_event_low_priority,
symbolicate_event_from_reprocessing_low_priority,
]
should_be_low_priority = should_demote_symbolication(project_id)
if is_low_priority != should_be_low_priority:
metrics.gauge("tasks.store.symbolicate_event.low_priority.wrong_queue", 1)
submit_symbolicate(
should_be_low_priority,
from_reprocessing,
cache_key,
event_id,
start_time,
data,
queue_switches + 1,
)
return
def _continue_to_process_event():
process_task = process_event_from_reprocessing if from_reprocessing else process_event
_do_process_event(
cache_key=cache_key,
start_time=start_time,
event_id=event_id,
process_task=process_task,
data=data,
data_has_changed=has_changed,
from_symbolicate=True,
)
symbolication_function = get_symbolication_function(data)
symbolication_function_name = getattr(symbolication_function, "__name__", "none")
if killswitch_matches_context(
"store.load-shed-symbolicate-event-projects",
{
"project_id": project_id,
"event_id": event_id,
"platform": data.get("platform") or "null",
"symbolication_function": symbolication_function_name,
},
):
return _continue_to_process_event()
has_changed = False
symbolication_start_time = time()
submission_ratio = options.get("symbolicate-event.low-priority.metrics.submission-rate")
submit_realtime_metrics = not from_reprocessing and random.random() < submission_ratio
if submit_realtime_metrics:
with sentry_sdk.start_span(op="tasks.store.symbolicate_event.low_priority.metrics.counter"):
timestamp = int(symbolication_start_time)
try:
realtime_metrics.increment_project_event_counter(project_id, timestamp)
except Exception as e:
sentry_sdk.capture_exception(e)
with sentry_sdk.start_span(op="tasks.store.symbolicate_event.symbolication") as span:
span.set_data("symbolication_function", symbolication_function_name)
with metrics.timer(
"tasks.store.symbolicate_event.symbolication",
tags={"symbolication_function": symbolication_function_name},
):
while True:
try:
with sentry_sdk.start_span(
op="tasks.store.symbolicate_event.%s" % symbolication_function_name
) as span:
symbolicated_data = symbolication_function(data)
span.set_data("symbolicated_data", bool(symbolicated_data))
if symbolicated_data:
data = symbolicated_data
has_changed = True
break
except RetrySymbolication as e:
if (
time() - symbolication_start_time
) > settings.SYMBOLICATOR_PROCESS_EVENT_WARN_TIMEOUT:
error_logger.warning(
"symbolicate.slow",
extra={"project_id": project_id, "event_id": event_id},
)
if (
time() - symbolication_start_time
) > settings.SYMBOLICATOR_PROCESS_EVENT_HARD_TIMEOUT:
# Do not drop event but actually continue with rest of pipeline
# (persisting unsymbolicated event)
metrics.incr(
"tasks.store.symbolicate_event.fatal",
tags={
"reason": "timeout",
"symbolication_function": symbolication_function_name,
},
)
error_logger.exception(
"symbolicate.failed.infinite_retry",
extra={"project_id": project_id, "event_id": event_id},
)
data.setdefault("_metrics", {})["flag.processing.error"] = True
data.setdefault("_metrics", {})["flag.processing.fatal"] = True
has_changed = True
break
else:
# sleep for `retry_after` but max 5 seconds and try again
metrics.incr(
"tasks.store.symbolicate_event.retry",
tags={"symbolication_function": symbolication_function_name},
)
sleep(min(e.retry_after, SYMBOLICATOR_MAX_RETRY_AFTER))
continue
except Exception:
metrics.incr(
"tasks.store.symbolicate_event.fatal",
tags={
"reason": "error",
"symbolication_function": symbolication_function_name,
},
)
error_logger.exception("tasks.store.symbolicate_event.symbolication")
data.setdefault("_metrics", {})["flag.processing.error"] = True
data.setdefault("_metrics", {})["flag.processing.fatal"] = True
has_changed = True
break
if submit_realtime_metrics:
with sentry_sdk.start_span(
op="tasks.store.symbolicate_event.low_priority.metrics.histogram"
):
symbolication_duration = int(time() - symbolication_start_time)
try:
realtime_metrics.increment_project_duration_counter(
project_id, timestamp, symbolication_duration
)
except Exception as e:
sentry_sdk.capture_exception(e)
# We cannot persist canonical types in the cache, so we need to
# downgrade this.
if isinstance(data, CANONICAL_TYPES):
data = dict(data.items())
if has_changed:
cache_key = event_processing_store.store(data)
return _continue_to_process_event()
def _do_save_event(cache_key=None,
data=None,
start_time=None,
event_id=None,
project_id=None,
**kwargs):
"""
Saves an event to the database.
"""
set_current_event_project(project_id)
from sentry.event_manager import EventManager, HashDiscarded
event_type = "none"
if cache_key and data is None:
with metrics.timer(
"tasks.store.do_save_event.get_cache") as metric_tags:
data = event_processing_store.get(cache_key)
if data is not None:
metric_tags["event_type"] = event_type = data.get(
"type") or "none"
with metrics.global_tags(event_type=event_type):
if data is not None:
data = CanonicalKeyDict(data)
if event_id is None and data is not None:
event_id = data["event_id"]
# only when we come from reprocessing we get a project_id sent into
# the task.
if project_id is None:
project_id = data.pop("project")
set_current_event_project(project_id)
# We only need to delete raw events for events that support
# reprocessing. If the data cannot be found we want to assume
# that we need to delete the raw event.
if not data or reprocessing.event_supports_reprocessing(data):
with metrics.timer("tasks.store.do_save_event.delete_raw_event"):
delete_raw_event(project_id, event_id, allow_hint_clear=True)
# This covers two cases: where data is None because we did not manage
# to fetch it from the default cache or the empty dictionary was
# stored in the default cache. The former happens if the event
# expired while being on the queue, the second happens on reprocessing
# if the raw event was deleted concurrently while we held on to
# it. This causes the node store to delete the data and we end up
# fetching an empty dict. We could in theory not invoke `save_event`
# in those cases but it's important that we always clean up the
# reprocessing reports correctly or they will screw up the UI. So
# to future proof this correctly we just handle this case here.
if not data:
metrics.incr("events.failed",
tags={
"reason": "cache",
"stage": "post"
},
skip_internal=False)
return
try:
with metrics.timer("tasks.store.do_save_event.event_manager.save"):
manager = EventManager(data)
# event.project.organization is populated after this statement.
manager.save(project_id,
assume_normalized=True,
start_time=start_time,
cache_key=cache_key)
# Put the updated event back into the cache so that post_process
# has the most recent data.
data = manager.get_data()
if isinstance(data, CANONICAL_TYPES):
data = dict(data.items())
with metrics.timer(
"tasks.store.do_save_event.write_processing_cache"):
event_processing_store.store(data)
except HashDiscarded:
# Delete the event payload from cache since it won't show up in post-processing.
if cache_key:
with metrics.timer("tasks.store.do_save_event.delete_cache"):
event_processing_store.delete_by_key(cache_key)
finally:
reprocessing2.mark_event_reprocessed(data)
if cache_key:
with metrics.timer(
"tasks.store.do_save_event.delete_attachment_cache"):
attachment_cache.delete(cache_key)
if start_time:
metrics.timing("events.time-to-process",
time() - start_time,
instance=data["platform"])
time_synthetic_monitoring_event(data, project_id, start_time)
def test_basic(task_runner, default_project, register_plugin, change_groups,
reset_snuba, change_stacktrace):
# Replace this with an int and nonlocal when we have Python 3
abs_count = []
def event_preprocessor(data):
tags = data.setdefault("tags", [])
assert all(not x or x[0] != "processing_counter" for x in tags)
tags.append(("processing_counter", "x{}".format(len(abs_count))))
abs_count.append(None)
if change_stacktrace and len(abs_count) > 0:
data["exception"] = {
"values": [{
"type": "ZeroDivisionError",
"stacktrace": {
"frames": [{
"function": "foo"
}]
}
}]
}
if change_groups:
data["fingerprint"] = [uuid.uuid4().hex]
else:
data["fingerprint"] = ["foo"]
return data
class ReprocessingTestPlugin(Plugin2):
def get_event_preprocessors(self, data):
return [event_preprocessor]
def is_enabled(self, project=None):
return True
register_plugin(globals(), ReprocessingTestPlugin)
mgr = EventManager(
data={
"timestamp": iso_format(before_now(seconds=1)),
"tags": [["key1", "value"], None, ["key2", "value"]],
},
project=default_project,
)
mgr.normalize()
data = mgr.get_data()
event_id = data["event_id"]
cache_key = event_processing_store.store(data)
def get_event_by_processing_counter(n):
return list(
eventstore.get_events(
eventstore.Filter(
project_ids=[default_project.id],
conditions=[["tags[processing_counter]", "=", n]],
)))
with task_runner(), Feature({"projects:reprocessing-v2": True}):
# factories.store_event would almost be suitable for this, but let's
# actually run through stacktrace processing once
preprocess_event(start_time=time(), cache_key=cache_key, data=data)
event = eventstore.get_event_by_id(default_project.id, event_id)
assert event.get_tag("processing_counter") == "x0"
assert not event.data.get("errors")
assert get_event_by_processing_counter("x0")[0].event_id == event.event_id
old_event = event
with task_runner(), Feature({"projects:reprocessing-v2": True}):
reprocess_group(default_project.id, event.group_id)
new_events = get_event_by_processing_counter("x1")
if not change_stacktrace:
assert not new_events
else:
(event, ) = new_events
# Assert original data is used
assert event.get_tag("processing_counter") == "x1"
assert not event.data.get("errors")
if change_groups:
assert event.group_id != old_event.group_id
else:
assert event.group_id == old_event.group_id
assert event.get_tag("original_event_id") == old_event.event_id
assert int(event.get_tag("original_group_id")) == old_event.group_id
def reprocess_event(project_id, event_id, start_time):
from sentry.tasks.store import preprocess_event_from_reprocessing
from sentry.ingest.ingest_consumer import CACHE_TIMEOUT
with sentry_sdk.start_span(op="reprocess_events.nodestore.get"):
node_id = Event.generate_node_id(project_id, event_id)
data = nodestore.get(node_id, subkey="unprocessed")
if data is None:
node_id = _generate_unprocessed_event_node_id(project_id=project_id, event_id=event_id)
data = nodestore.get(node_id)
with sentry_sdk.start_span(op="reprocess_events.eventstore.get"):
event = eventstore.get_event_by_id(project_id, event_id)
if event is None:
logger.error(
"reprocessing2.event.not_found", extra={"project_id": project_id, "event_id": event_id}
)
return
if data is None:
logger.error(
"reprocessing2.reprocessing_nodestore.not_found",
extra={"project_id": project_id, "event_id": event_id},
)
# We have no real data for reprocessing. We assume this event goes
# straight to save_event, and hope that the event data can be
# reingested like that. It's better than data loss.
#
# XXX: Ideally we would run a "save-lite" for this that only updates
# the group ID in-place. Like a snuba merge message.
data = dict(event.data)
# Step 1: Fix up the event payload for reprocessing and put it in event
# cache/event_processing_store
set_path(data, "contexts", "reprocessing", "original_issue_id", value=event.group_id)
cache_key = event_processing_store.store(data)
# Step 2: Copy attachments into attachment cache
queryset = models.EventAttachment.objects.filter(project_id=project_id, event_id=event_id)
files = {f.id: f for f in models.File.objects.filter(id__in=[ea.file_id for ea in queryset])}
attachment_objects = []
for attachment_id, attachment in enumerate(queryset):
with sentry_sdk.start_span(op="reprocess_event._copy_attachment_into_cache") as span:
span.set_data("attachment_id", attachment.id)
attachment_objects.append(
_copy_attachment_into_cache(
attachment_id=attachment_id,
attachment=attachment,
file=files[attachment.file_id],
cache_key=cache_key,
cache_timeout=CACHE_TIMEOUT,
)
)
if attachment_objects:
with sentry_sdk.start_span(op="reprocess_event.set_attachment_meta"):
attachment_cache.set(cache_key, attachments=attachment_objects, timeout=CACHE_TIMEOUT)
preprocess_event_from_reprocessing(
cache_key=cache_key, start_time=start_time, event_id=event_id
)
def reprocess_event(project_id, event_id, start_time):
from sentry.event_manager import set_tag
from sentry.tasks.store import preprocess_event_from_reprocessing
from sentry.ingest.ingest_consumer import CACHE_TIMEOUT
# Take unprocessed data from old event and save it as unprocessed data
# under a new event ID. The second step happens in pre-process. We could
# save the "original event ID" instead and get away with writing less to
# nodestore, but doing it this way makes the logic slightly simpler.
node_id = _generate_unprocessed_event_node_id(project_id=project_id,
event_id=event_id)
with sentry_sdk.start_span(op="reprocess_events.nodestore.get"):
data = nodestore.get(node_id)
with sentry_sdk.start_span(op="reprocess_events.eventstore.get"):
event = eventstore.get_event_by_id(project_id, event_id)
if event is None:
logger.error("reprocessing2.event.not_found",
extra={
"project_id": project_id,
"event_id": event_id
})
return
if data is None:
logger.error(
"reprocessing2.reprocessing_nodestore.not_found",
extra={
"project_id": project_id,
"event_id": event_id
},
)
# We have no real data for reprocessing. We assume this event goes
# straight to save_event, and hope that the event data can be
# reingested like that. It's better than data loss.
#
# XXX: Ideally we would run a "save-lite" for this that only updates
# the group ID in-place. Like a snuba merge message.
data = dict(event.data)
# Step 1: Fix up the event payload for reprocessing and put it in event
# cache/event_processing_store
set_tag(data, "original_group_id", event.group_id)
cache_key = event_processing_store.store(data)
# Step 2: Copy attachments into attachment cache
queryset = models.EventAttachment.objects.filter(
project_id=project_id, event_id=event_id).select_related("file")
attachment_objects = []
for attachment_id, attachment in enumerate(queryset):
with sentry_sdk.start_span(
op="reprocess_event._copy_attachment_into_cache") as span:
span.set_data("attachment_id", attachment.id)
attachment_objects.append(
_copy_attachment_into_cache(
attachment_id=attachment_id,
attachment=attachment,
cache_key=cache_key,
cache_timeout=CACHE_TIMEOUT,
))
if attachment_objects:
with sentry_sdk.start_span(op="reprocess_event.set_attachment_meta"):
attachment_cache.set(cache_key,
attachments=attachment_objects,
timeout=CACHE_TIMEOUT)
preprocess_event_from_reprocessing(cache_key=cache_key,
start_time=start_time,
event_id=event_id)
def reprocess_event(project_id, event_id, start_time):
from sentry.ingest.ingest_consumer import CACHE_TIMEOUT
from sentry.tasks.store import preprocess_event_from_reprocessing
reprocessable_event = pull_event_data(project_id, event_id)
data = reprocessable_event.data
event = reprocessable_event.event
attachments = reprocessable_event.attachments
# Step 1: Fix up the event payload for reprocessing and put it in event
# cache/event_processing_store
set_path(data,
"contexts",
"reprocessing",
"original_issue_id",
value=event.group_id)
set_path(data,
"contexts",
"reprocessing",
"original_primary_hash",
value=event.get_primary_hash())
cache_key = event_processing_store.store(data)
# Step 2: Copy attachments into attachment cache. Note that we can only
# consider minidumps because filestore just stays as-is after reprocessing
# (we simply update group_id on the EventAttachment models in post_process)
attachment_objects = []
files = {
f.id: f
for f in models.File.objects.filter(
id__in=[ea.file_id for ea in attachments])
}
for attachment_id, attachment in enumerate(attachments):
with sentry_sdk.start_span(
op="reprocess_event._copy_attachment_into_cache") as span:
span.set_data("attachment_id", attachment.id)
attachment_objects.append(
_copy_attachment_into_cache(
attachment_id=attachment_id,
attachment=attachment,
file=files[attachment.file_id],
cache_key=cache_key,
cache_timeout=CACHE_TIMEOUT,
))
if attachment_objects:
with sentry_sdk.start_span(op="reprocess_event.set_attachment_meta"):
attachment_cache.set(cache_key,
attachments=attachment_objects,
timeout=CACHE_TIMEOUT)
preprocess_event_from_reprocessing(
cache_key=cache_key,
start_time=start_time,
event_id=event_id,
data=data,
)
