def _normalize_impl(self, project_id=None):
if self._project and project_id and project_id != self._project.id:
raise RuntimeError(
"Initialized EventManager with one project ID and called save() with another one"
)
if self._normalized:
raise RuntimeError("Already normalized")
self._normalized = True
from sentry_relay.processing import StoreNormalizer
rust_normalizer = StoreNormalizer(
project_id=self._project.id if self._project else project_id,
client_ip=self._client_ip,
client=self._auth.client if self._auth else None,
key_id=six.text_type(self._key.id) if self._key else None,
grouping_config=self._grouping_config,
protocol_version=six.text_type(self.version) if self.version is not None else None,
is_renormalize=self._is_renormalize,
remove_other=self._remove_other,
normalize_user_agent=True,
sent_at=self.sent_at.isoformat() if self.sent_at is not None else None,
**DEFAULT_STORE_NORMALIZER_ARGS
)
self._data = CanonicalKeyDict(rust_normalizer.normalize_event(dict(self._data)))
def __init__(self, data, skip_renormalization=False, **kwargs):
is_renormalized = isinstance(data, EventDict) or (
isinstance(data, NodeData) and isinstance(data.data, EventDict)
)
if not skip_renormalization and not is_renormalized:
normalizer = StoreNormalizer(is_renormalize=True, enable_trimming=False)
data = normalizer.normalize_event(dict(data))
CanonicalKeyDict.__init__(self, data, **kwargs)
def _normalize_impl(self):
if self._normalized:
raise RuntimeError("Already normalized")
self._normalized = True
from sentry_relay.processing import StoreNormalizer
rust_normalizer = StoreNormalizer(
project_id=self._project.id if self._project else None,
client_ip=self._client_ip,
client=self._auth.client if self._auth else None,
key_id=six.text_type(self._key.id) if self._key else None,
grouping_config=self._grouping_config,
protocol_version=six.text_type(self.version)
if self.version is not None else None,
is_renormalize=self._is_renormalize,
remove_other=self._remove_other,
normalize_user_agent=True,
**DEFAULT_STORE_NORMALIZER_ARGS)
self._data = CanonicalKeyDict(
rust_normalizer.normalize_event(dict(self._data)))
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_process_event(cache_key,
start_time,
event_id,
process_task,
data=None):
from sentry.plugins.base import plugins
if data is None:
data = default_cache.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"]
event_id = data["event_id"]
with configure_scope() as scope:
scope.set_tag("project", project_id)
has_changed = False
# Fetch the reprocessing revision
reprocessing_rev = reprocessing.get_reprocessing_revision(project_id)
try:
# Event enhancers. These run before anything else.
for plugin in plugins.all(version=2):
enhancers = safe_execute(plugin.get_event_enhancers, data=data)
for enhancer in enhancers or ():
enhanced = safe_execute(
enhancer, data, _passthrough_errors=(RetrySymbolication, ))
if enhanced:
data = enhanced
has_changed = True
# Stacktrace based event processors.
new_data = process_stacktraces(data)
if new_data is not None:
has_changed = True
data = new_data
except RetrySymbolication as e:
if start_time and (time() - start_time) > 120:
error_logger.warning("process.slow",
extra={
"project_id": project_id,
"event_id": event_id
})
if start_time and (time() - start_time) > 3600:
# Do not drop event but actually continue with rest of pipeline
# (persisting unsymbolicated event)
error_logger.exception(
"process.failed.infinite_retry",
extra={
"project_id": project_id,
"event_id": event_id
},
)
else:
retry_process_event.apply_async(
args=(),
kwargs={
"process_task_name": process_task.__name__,
"task_kwargs": {
"cache_key": cache_key,
"event_id": event_id,
"start_time": start_time,
},
},
countdown=e.retry_after,
)
return
# TODO(dcramer): ideally we would know if data changed by default
# Default event processors.
for plugin in plugins.all(version=2):
processors = safe_execute(plugin.get_event_preprocessors,
data=data,
_with_transaction=False)
for processor in processors or ():
result = safe_execute(processor, data)
if result:
data = result
has_changed = True
assert data[
"project"] == project_id, "Project cannot be mutated by preprocessor"
project = Project.objects.get_from_cache(id=project_id)
# 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 outselves again. This happens when the reprocessing
# revision changed while we were processing.
from_reprocessing = process_task is process_event_from_reprocessing
submit_process(project, from_reprocessing, cache_key, event_id,
start_time, data)
process_task.delay(cache_key,
start_time=start_time,
event_id=event_id)
return
default_cache.set(cache_key, data, 3600)
submit_save_event(project, cache_key, event_id, start_time, data)
def normalize_event(event, project_id):
normalizer = StoreNormalizer(project_id=project_id, )
return normalizer.normalize_event(event)
def _do_process_event(cache_key, start_time, event_id, process_task, data=None):
from sentry.plugins.base import plugins
if data is None:
data = default_cache.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"]
event_id = data["event_id"]
project = Project.objects.get_from_cache(id=project_id)
with configure_scope() as scope:
scope.set_tag("project", project_id)
has_changed = False
# Fetch the reprocessing revision
reprocessing_rev = reprocessing.get_reprocessing_revision(project_id)
try:
# Event enhancers. These run before anything else.
for plugin in plugins.all(version=2):
enhancers = safe_execute(plugin.get_event_enhancers, data=data)
for enhancer in enhancers or ():
enhanced = safe_execute(enhancer, data, _passthrough_errors=(RetrySymbolication,))
if enhanced:
data = enhanced
has_changed = True
# Stacktrace based event processors.
new_data = process_stacktraces(data)
if new_data is not None:
has_changed = True
data = new_data
except RetrySymbolication as e:
if start_time and (time() - start_time) > settings.SYMBOLICATOR_PROCESS_EVENT_WARN_TIMEOUT:
error_logger.warning(
"process.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(
"process.failed.infinite_retry",
extra={"project_id": project_id, "event_id": event_id},
)
else:
retry_process_event.apply_async(
args=(),
kwargs={
"process_task_name": process_task.__name__,
"task_kwargs": {
"cache_key": cache_key,
"event_id": event_id,
"start_time": start_time,
},
},
countdown=e.retry_after,
)
return
# Second round of datascrubbing after stacktrace and language-specific
# processing. First round happened as part of ingest.
#
# We assume that all potential PII is produced as part of stacktrace
# processors and event enhancers.
#
# We assume that plugins for eg sessionstack (running via
# `plugin.get_event_preprocessors`) are not producing data that should be
# PII-stripped, ever.
#
# XXX(markus): Javascript event error translation is happening after this block
# because it uses `get_event_preprocessors` instead of
# `get_event_enhancers`, possibly move?
if has_changed and features.has(
"organizations:datascrubbers-v2", project.organization, actor=None
):
with metrics.timer("tasks.store.datascrubbers.scrub"):
project_config = get_project_config(project)
new_data = safe_execute(scrub_data, project_config=project_config, 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):
processors = safe_execute(
plugin.get_event_preprocessors, data=data, _with_transaction=False
)
for processor in processors or ():
result = safe_execute(processor, data)
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 outselves again. This happens when the reprocessing
# revision changed while we were processing.
from_reprocessing = process_task is process_event_from_reprocessing
submit_process(project, from_reprocessing, cache_key, event_id, start_time, data)
process_task.delay(cache_key, start_time=start_time, event_id=event_id)
return
default_cache.set(cache_key, data, 3600)
submit_save_event(project, cache_key, event_id, start_time, data)
def _do_process_event(
cache_key,
start_time,
event_id,
process_task,
data=None,
data_has_changed=None,
new_process_behavior=None,
):
from sentry.plugins.base import plugins
if data is None:
data = default_cache.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_project(project_id)
event_id = data["event_id"]
project = Project.objects.get_from_cache(id=project_id)
has_changed = bool(data_has_changed)
new_process_behavior = bool(new_process_behavior)
metrics.incr("tasks.store.process_event.new_process_behavior",
tags={"value": new_process_behavior})
# Fetch the reprocessing revision
reprocessing_rev = reprocessing.get_reprocessing_revision(project_id)
try:
if not new_process_behavior:
# Event enhancers. These run before anything else.
for plugin in plugins.all(version=2):
with metrics.timer("tasks.store.process_event.enhancers",
tags={"plugin": plugin.slug}):
enhancers = safe_execute(plugin.get_event_enhancers,
data=data)
for enhancer in enhancers or ():
enhanced = safe_execute(
enhancer,
data,
_passthrough_errors=(RetrySymbolication, ))
if enhanced:
data = enhanced
has_changed = True
# Stacktrace based event processors.
with metrics.timer("tasks.store.process_event.stacktraces"):
new_data = process_stacktraces(data)
if new_data is not None:
has_changed = True
data = new_data
except RetrySymbolication as e:
if start_time and (
time() -
start_time) > settings.SYMBOLICATOR_PROCESS_EVENT_WARN_TIMEOUT:
error_logger.warning("process.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(
"process.failed.infinite_retry",
extra={
"project_id": project_id,
"event_id": event_id
},
)
else:
retry_process_event.apply_async(
args=(),
kwargs={
"process_task_name": process_task.__name__,
"task_kwargs": {
"cache_key": cache_key,
"event_id": event_id,
"start_time": start_time,
},
},
countdown=e.retry_after,
)
return
# 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 and `get_event_enhancers`.
#
# 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")
and features.has("organizations:datascrubbers-v2",
project.organization,
actor=None)):
with metrics.timer("tasks.store.datascrubbers.scrub"):
project_config = get_project_config(project)
new_data = safe_execute(scrub_data,
project_config=project_config,
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 metrics.timer("tasks.store.process_event.preprocessors",
tags={"plugin": plugin.slug}):
processors = safe_execute(plugin.get_event_preprocessors,
data=data,
_with_transaction=False)
for processor in processors or ():
result = safe_execute(processor, data)
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 outselves 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
default_cache.set(cache_key, data, 3600)
submit_save_event(project, cache_key, event_id, start_time, data)