closeyourit-ruby 0.10.0 → 0.10.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- checksums.yaml +4 -4
- data/lib/closeyourit/background_worker.rb +14 -14
- data/lib/closeyourit/breadcrumb.rb +2 -2
- data/lib/closeyourit/breadcrumb_buffer.rb +2 -2
- data/lib/closeyourit/client.rb +24 -24
- data/lib/closeyourit/configuration.rb +102 -102
- data/lib/closeyourit/event.rb +6 -6
- data/lib/closeyourit/events/error_event.rb +16 -16
- data/lib/closeyourit/events/job_metric_event.rb +8 -8
- data/lib/closeyourit/events/log_event.rb +17 -17
- data/lib/closeyourit/events/message_event.rb +4 -4
- data/lib/closeyourit/events/performance_issue_event.rb +4 -4
- data/lib/closeyourit/events/slow_method_event.rb +6 -6
- data/lib/closeyourit/events/slow_query_event.rb +4 -4
- data/lib/closeyourit/instrumenter.rb +6 -6
- data/lib/closeyourit/line_cache.rb +4 -4
- data/lib/closeyourit/log_buffer.rb +10 -10
- data/lib/closeyourit/log_device.rb +18 -18
- data/lib/closeyourit/monitor.rb +2 -2
- data/lib/closeyourit/performance/request_profile.rb +5 -5
- data/lib/closeyourit/performance/rollup.rb +3 -3
- data/lib/closeyourit/rails/active_job_extension.rb +33 -31
- data/lib/closeyourit/rails/capture_exceptions.rb +3 -3
- data/lib/closeyourit/rails/error_subscriber.rb +4 -4
- data/lib/closeyourit/rails/log_broadcast.rb +12 -12
- data/lib/closeyourit/rails/net_http_patch.rb +22 -22
- data/lib/closeyourit/rails/query_source.rb +3 -3
- data/lib/closeyourit/rails/railtie.rb +32 -52
- data/lib/closeyourit/rails/request_body.rb +7 -7
- data/lib/closeyourit/rails/request_context.rb +27 -27
- data/lib/closeyourit/scope.rb +29 -29
- data/lib/closeyourit/scrubber.rb +47 -48
- data/lib/closeyourit/sidekiq/error_handler.rb +2 -2
- data/lib/closeyourit/sidekiq/job_metrics_middleware.rb +10 -11
- data/lib/closeyourit/stats.rb +8 -8
- data/lib/closeyourit/subscribers/job_performance.rb +30 -19
- data/lib/closeyourit/subscribers/request_performance.rb +4 -4
- data/lib/closeyourit/subscribers/slow_query.rb +42 -20
- data/lib/closeyourit/trace_context.rb +22 -22
- data/lib/closeyourit/transport.rb +22 -22
- data/lib/closeyourit/usage_registry.rb +17 -16
- data/lib/closeyourit/version.rb +1 -1
- data/lib/closeyourit-ruby.rb +125 -125
- metadata +1 -1
|
@@ -5,16 +5,16 @@ require_relative "../trace_context"
|
|
|
5
5
|
|
|
6
6
|
module CloseYourIt
|
|
7
7
|
module Rails
|
|
8
|
-
# Rack middleware:
|
|
9
|
-
#
|
|
10
|
-
# `CaptureExceptions` (insert_before)
|
|
11
|
-
# Rack
|
|
8
|
+
# Rack middleware: populates the Scope with the request's HTTP context (method/url/headers)
|
|
9
|
+
# so the error event captured downstream knows "on which page" it happened. It must wrap
|
|
10
|
+
# `CaptureExceptions` (insert_before) to be already populated when the exception bubbles up.
|
|
11
|
+
# Pure Rack (reads `env`, no Rails dependency) → testable in isolation.
|
|
12
12
|
class RequestContext
|
|
13
|
-
#
|
|
13
|
+
# Headers without the HTTP_ prefix in the Rack env.
|
|
14
14
|
RAW_HEADERS = %w[CONTENT_TYPE CONTENT_LENGTH].freeze
|
|
15
15
|
|
|
16
|
-
#
|
|
17
|
-
#
|
|
16
|
+
# Correlation cookie written by the browser SDK (session replay): opaque id of the
|
|
17
|
+
# replay session, carried by every same-origin request.
|
|
18
18
|
REPLAY_COOKIE = "cyi_replay"
|
|
19
19
|
|
|
20
20
|
def initialize(app)
|
|
@@ -23,19 +23,19 @@ module CloseYourIt
|
|
|
23
23
|
|
|
24
24
|
def call(env)
|
|
25
25
|
if enabled?
|
|
26
|
-
#
|
|
26
|
+
# W3C trace context (only with opt-in propagation): adopted from the incoming header or generated.
|
|
27
27
|
context = trace_context_for(env)
|
|
28
28
|
CloseYourIt::Scope.current.trace_context = context
|
|
29
|
-
# trace_id
|
|
30
|
-
#
|
|
31
|
-
#
|
|
29
|
+
# trace_id always (log↔error correlation), even with capture_request OFF. With a W3C trace
|
|
30
|
+
# context the events' trace_id IS the W3C trace-id → the error/metric lines up with the
|
|
31
|
+
# distributed trace propagated downstream (trace_id mapping, CYRB-15); otherwise request_id as before.
|
|
32
32
|
CloseYourIt::Scope.current.trace_id = context ? context.trace_id : trace_id_for(env)
|
|
33
|
-
#
|
|
34
|
-
# → contexts.replay.replay_id
|
|
33
|
+
# Server error ↔ session replay correlation: the id from the cookie lands on the scope
|
|
34
|
+
# → the event's contexts.replay.replay_id (same spot as the JS path).
|
|
35
35
|
CloseYourIt::Scope.current.replay_session_id = replay_session_id_for(env)
|
|
36
36
|
if CloseYourIt.configuration.capture_request
|
|
37
37
|
CloseYourIt::Scope.current.request = build_request(env)
|
|
38
|
-
#
|
|
38
|
+
# Reference to the env for LAZY body extraction (request.data) once the event is built.
|
|
39
39
|
CloseYourIt::Scope.current.rack_env = env
|
|
40
40
|
end
|
|
41
41
|
end
|
|
@@ -52,10 +52,10 @@ module CloseYourIt
|
|
|
52
52
|
false
|
|
53
53
|
end
|
|
54
54
|
|
|
55
|
-
#
|
|
56
|
-
#
|
|
57
|
-
#
|
|
58
|
-
#
|
|
55
|
+
# W3C trace context of the request, ONLY with opt-in propagation on (otherwise nil → historical
|
|
56
|
+
# behavior unchanged). Adopts a valid incoming traceparent/tracestate (distributed correlation),
|
|
57
|
+
# otherwise generates a root: so services that ORIGINATE traffic also start with a W3C trace-id
|
|
58
|
+
# that can be propagated downstream. A malformed traceparent is ignored → a root is generated.
|
|
59
59
|
def trace_context_for(env)
|
|
60
60
|
return nil unless CloseYourIt.configuration.propagate_trace_context
|
|
61
61
|
|
|
@@ -63,28 +63,28 @@ module CloseYourIt
|
|
|
63
63
|
CloseYourIt::TraceContext.generate
|
|
64
64
|
end
|
|
65
65
|
|
|
66
|
-
#
|
|
67
|
-
#
|
|
66
|
+
# Reuses the Rails/Rack request id when present (same correlation as the application logs),
|
|
67
|
+
# otherwise generates one.
|
|
68
68
|
def trace_id_for(env)
|
|
69
69
|
env["action_dispatch.request_id"] ||
|
|
70
70
|
env["HTTP_X_REQUEST_ID"].to_s.split(",").first&.strip.then { |id| id&.empty? ? nil : id } ||
|
|
71
71
|
SecureRandom.uuid
|
|
72
72
|
end
|
|
73
73
|
|
|
74
|
-
#
|
|
74
|
+
# Extracts `cyi_replay` from the request cookie (nil if absent). Pure Rack.
|
|
75
75
|
def replay_session_id_for(env)
|
|
76
76
|
cookie = env["HTTP_COOKIE"]
|
|
77
77
|
return nil if cookie.nil? || cookie.empty?
|
|
78
78
|
|
|
79
|
-
# `rack/utils`
|
|
80
|
-
#
|
|
81
|
-
#
|
|
79
|
+
# `rack/utils` loaded lazily: the middleware runs only inside a Rack app (where Rack is surely
|
|
80
|
+
# present), so `require "closeyourit-ruby"` does not force Rack on non-web apps (CLI/workers)
|
|
81
|
+
# and the gem stays installable without declaring `rack` as a dependency (CYRB-13).
|
|
82
82
|
require "rack/utils"
|
|
83
83
|
Rack::Utils.parse_cookies_header(cookie)[REPLAY_COOKIE].to_s.then { |id| id.empty? ? nil : id }
|
|
84
84
|
end
|
|
85
85
|
|
|
86
|
-
#
|
|
87
|
-
# Authorization/Cookie). query_string + IP
|
|
86
|
+
# Sentry `request` shape. URL without query string; headers only from the allowlist (never
|
|
87
|
+
# Authorization/Cookie). query_string + IP only with send_pii (opt-in).
|
|
88
88
|
def build_request(env)
|
|
89
89
|
request = {
|
|
90
90
|
"method" => env["REQUEST_METHOD"],
|
|
@@ -101,7 +101,7 @@ module CloseYourIt
|
|
|
101
101
|
|
|
102
102
|
request
|
|
103
103
|
rescue StandardError
|
|
104
|
-
#
|
|
104
|
+
# Telemetry must never disturb the host request.
|
|
105
105
|
nil
|
|
106
106
|
end
|
|
107
107
|
|
data/lib/closeyourit/scope.rb
CHANGED
|
@@ -4,29 +4,29 @@ require_relative "breadcrumb_buffer"
|
|
|
4
4
|
require_relative "performance/request_profile"
|
|
5
5
|
|
|
6
6
|
module CloseYourIt
|
|
7
|
-
#
|
|
8
|
-
# user/tags/extra/contexts/request.
|
|
9
|
-
# →
|
|
7
|
+
# Per-request (or per-job) context isolated per execution context (Fiber storage):
|
|
8
|
+
# user/tags/extra/contexts/request. Read by ErrorEvent#to_h on the calling thread (synchronous)
|
|
9
|
+
# → the send worker never sees it and the scope does not leak between requests.
|
|
10
10
|
class Scope
|
|
11
11
|
STORAGE_KEY = :__closeyourit_scope
|
|
12
12
|
|
|
13
13
|
class << self
|
|
14
|
-
# Scope
|
|
15
|
-
#
|
|
16
|
-
# `Thread.current` (thread-local
|
|
14
|
+
# Scope of the current execution context. Uses `ActiveSupport::IsolatedExecutionState` when
|
|
15
|
+
# present (honors isolation_level: thread on Puma, fiber on Falcon), otherwise
|
|
16
|
+
# `Thread.current` (pure thread-local, NOT inherited by child threads → no bleed).
|
|
17
17
|
def current
|
|
18
18
|
store[STORAGE_KEY] ||= new
|
|
19
19
|
end
|
|
20
20
|
|
|
21
|
-
#
|
|
22
|
-
#
|
|
23
|
-
#
|
|
21
|
+
# Reinstalls a previously saved scope. Used by `after_discard` (ActiveJob 7.1+) to get back the
|
|
22
|
+
# scope enriched during `perform` — tags/contexts/breadcrumbs, queries included — after the
|
|
23
|
+
# end-of-`perform` reset detached it from storage (CYRB-19). Symmetric to `reset!`.
|
|
24
24
|
def current=(scope)
|
|
25
25
|
store[STORAGE_KEY] = scope
|
|
26
26
|
end
|
|
27
27
|
|
|
28
|
-
#
|
|
29
|
-
# thread
|
|
28
|
+
# Clears the current scope — called in `ensure` by middleware and jobs (on Puma the
|
|
29
|
+
# thread is reused: without a reset the scope would leak into the next request).
|
|
30
30
|
def reset!
|
|
31
31
|
store[STORAGE_KEY] = nil
|
|
32
32
|
end
|
|
@@ -73,8 +73,8 @@ module CloseYourIt
|
|
|
73
73
|
@breadcrumbs.add(breadcrumb)
|
|
74
74
|
end
|
|
75
75
|
|
|
76
|
-
#
|
|
77
|
-
#
|
|
76
|
+
# Per-request performance profile (queries + external HTTP). Lazy: created on first access,
|
|
77
|
+
# reset by #clear at the end of the request. Subscribers::RequestPerformance computes the verdict.
|
|
78
78
|
def performance_profile
|
|
79
79
|
@performance_profile ||= Performance::RequestProfile.new
|
|
80
80
|
end
|
|
@@ -88,17 +88,17 @@ module CloseYourIt
|
|
|
88
88
|
@rack_env = nil
|
|
89
89
|
@trace_id = nil
|
|
90
90
|
@replay_session_id = nil
|
|
91
|
-
#
|
|
92
|
-
#
|
|
91
|
+
# W3C trace context of the request (CloseYourIt::TraceContext): populated only when opt-in
|
|
92
|
+
# propagation is on, consumed by the Net::HTTP patch for outgoing headers.
|
|
93
93
|
@trace_context = nil
|
|
94
94
|
@breadcrumbs = BreadcrumbBuffer.new(CloseYourIt.configuration.max_breadcrumbs)
|
|
95
95
|
@performance_profile = nil
|
|
96
96
|
end
|
|
97
97
|
|
|
98
|
-
#
|
|
99
|
-
#
|
|
100
|
-
#
|
|
101
|
-
#
|
|
98
|
+
# Non-empty subset in Sentry event shape (user/tags/extra/contexts/request), merged into the
|
|
99
|
+
# payload by ErrorEvent#to_h. tags/extra/contexts go through the Scrubber (recursive denylist by
|
|
100
|
+
# key): the backend does NOT scrub them again (Errors::Ingest::Normalize keeps them verbatim),
|
|
101
|
+
# so this is the only safety net against sensitive keys there — R2.
|
|
102
102
|
def to_event_hash
|
|
103
103
|
{
|
|
104
104
|
"user" => serialize_user,
|
|
@@ -112,26 +112,26 @@ module CloseYourIt
|
|
|
112
112
|
|
|
113
113
|
private
|
|
114
114
|
|
|
115
|
-
# contexts
|
|
116
|
-
#
|
|
117
|
-
#
|
|
115
|
+
# User contexts + `replay.replay_id` (session replay) when present on the scope: the backend
|
|
116
|
+
# reads contexts.replay.replay_id to link the server error to the video (same spot as the JS
|
|
117
|
+
# path). replay/replay_id are not sensitive keys → they survive the Scrubber.
|
|
118
118
|
def contexts_payload
|
|
119
119
|
merged = @contexts.dup
|
|
120
120
|
merged["replay"] = { "replay_id" => @replay_session_id } if @replay_session_id
|
|
121
121
|
scrub(presence(merged))
|
|
122
122
|
end
|
|
123
123
|
|
|
124
|
-
#
|
|
125
|
-
#
|
|
126
|
-
#
|
|
124
|
+
# Redacts the values of sensitive keys preserving the structure (e.g. contexts.runtime stays
|
|
125
|
+
# intact, only values under sensitive keys become [FILTERED]). Reuses the configuration's
|
|
126
|
+
# Scrubber, the same path as breadcrumb.data and log attributes.
|
|
127
127
|
def scrub(hash)
|
|
128
128
|
return hash if hash.nil?
|
|
129
129
|
|
|
130
130
|
Scrubber.new(CloseYourIt.configuration).filter_params(hash)
|
|
131
131
|
end
|
|
132
132
|
|
|
133
|
-
# Request context + body params (`request.data`)
|
|
134
|
-
#
|
|
133
|
+
# Request context + body params (`request.data`) extracted LAZILY here — i.e. only when an
|
|
134
|
+
# event is actually built, never on the request's happy path.
|
|
135
135
|
def request_payload
|
|
136
136
|
return nil if @request.nil?
|
|
137
137
|
|
|
@@ -148,8 +148,8 @@ module CloseYourIt
|
|
|
148
148
|
nil
|
|
149
149
|
end
|
|
150
150
|
|
|
151
|
-
# `user.id`
|
|
152
|
-
#
|
|
151
|
+
# `user.id` always; email/ip_address/username only with `send_pii` (the backend strips them
|
|
152
|
+
# anyway — defense in depth).
|
|
153
153
|
def serialize_user
|
|
154
154
|
return nil if @user.empty?
|
|
155
155
|
return @user if CloseYourIt.configuration.send_pii
|
data/lib/closeyourit/scrubber.rb
CHANGED
|
@@ -1,73 +1,72 @@
|
|
|
1
1
|
# frozen_string_literal: true
|
|
2
2
|
|
|
3
3
|
module CloseYourIt
|
|
4
|
-
#
|
|
5
|
-
# Privacy-by-default —
|
|
4
|
+
# PII removal from payloads: sensitive key filter, SQL normalization, message scrubbing.
|
|
5
|
+
# Privacy-by-default — see PDR §9.
|
|
6
6
|
class Scrubber
|
|
7
7
|
FILTERED = "[FILTERED]"
|
|
8
8
|
|
|
9
|
-
#
|
|
10
|
-
# SENSITIVE_KEY
|
|
11
|
-
# Errors/Logs::Ingest::Normalize::SENSITIVE_KEY.
|
|
9
|
+
# Key tokens that are always redacted (substring match, normalized). Aligned with the
|
|
10
|
+
# SENSITIVE_KEY regex of the backend and the Dart client (client-side parity) — see
|
|
11
|
+
# Errors/Logs::Ingest::Normalize::SENSITIVE_KEY. Covers credentials/secrets and PII:
|
|
12
12
|
# /pass|secret|token|api[_-]?key|apikey|authorization|cookie|csrf|credit|card|cvv|ssn|iban|
|
|
13
13
|
# email|phone|telephone|mobile|dob|birth|passport|bearer|session|pin|pan/i
|
|
14
|
-
# `pass`
|
|
15
|
-
# `credit`+`card`
|
|
16
|
-
# date_of_birth.
|
|
17
|
-
#
|
|
18
|
-
#
|
|
19
|
-
# CYRB-3:
|
|
14
|
+
# `pass` covers password/passwd/pass_code/passkey/passphrase; `cookie` covers set-cookie;
|
|
15
|
+
# `credit`+`card` cover credit_card; `phone` covers telephone/smartphone; `birth` covers
|
|
16
|
+
# date_of_birth. Substring match → favors over-redaction (privacy-by-default):
|
|
17
|
+
# e.g. `company_name` (contains `pan`) or `shipping` (contains `pin`) get redacted — that is
|
|
18
|
+
# acceptable, better to redact too much than to leak PII.
|
|
19
|
+
# CYRB-3: the list omitted email/phone/dob → slow query binds leaked them into the panel.
|
|
20
20
|
DENYLIST = %w[
|
|
21
21
|
pass secret token api_key apikey authorization
|
|
22
22
|
cookie csrf credit card cvv ssn iban
|
|
23
23
|
email phone telephone mobile dob birth passport bearer session pin pan
|
|
24
24
|
].freeze
|
|
25
25
|
|
|
26
|
-
# CYRB-23 —
|
|
27
|
-
# `filter_params`/`filter_value`
|
|
28
|
-
# DENYLIST);
|
|
29
|
-
#
|
|
30
|
-
#
|
|
26
|
+
# CYRB-23 — BUILT-IN redaction of authentication credentials in free text.
|
|
27
|
+
# `filter_params`/`filter_value` cover STRUCTURED headers (the `authorization` key is in
|
|
28
|
+
# DENYLIST); an `Authorization: Bearer <token>` pasted into `exception.message` or into a log
|
|
29
|
+
# line is text, not a key, and before this ticket it was sent in clear because `scrub_message`
|
|
30
|
+
# applied only the user-configured patterns (default `[]`).
|
|
31
31
|
#
|
|
32
|
-
#
|
|
33
|
-
# CYPY-15):
|
|
34
|
-
# `Authorization: Bearer [FILTERED]`.
|
|
35
|
-
#
|
|
32
|
+
# Shape of the redacted text (decided here once for all SDKs — twins CYDA-26/CYJS-38/
|
|
33
|
+
# CYPY-15): key and scheme stay readable, ONLY the credential disappears →
|
|
34
|
+
# `Authorization: Bearer [FILTERED]`. The message stays useful for diagnosis (you see THAT
|
|
35
|
+
# there was a header and with which scheme) without carrying the secret.
|
|
36
36
|
AUTH_SCHEME = /(?i:Bearer|Basic|Token)/
|
|
37
37
|
|
|
38
|
-
# `token68` (RFC 7235)
|
|
39
|
-
#
|
|
40
|
-
#
|
|
38
|
+
# `token68` (RFC 7235) with trailing `=` padding: the shape of a real credential. Minimum 4
|
|
39
|
+
# characters — a `Basic` with short credentials is short (`dTpw` = `u:p`); filtering out prose
|
|
40
|
+
# is NOT_PROSE's job, not the length's.
|
|
41
41
|
CREDENTIAL_TOKEN = %r{[A-Za-z0-9\-._~+/]{4,}={0,2}}
|
|
42
42
|
|
|
43
|
-
#
|
|
44
|
-
#
|
|
45
|
-
# `
|
|
46
|
-
#
|
|
47
|
-
#
|
|
48
|
-
# (`dTpw`, base64)
|
|
49
|
-
#
|
|
50
|
-
# Trade-off accettato: un CamelCase subito dopo lo schema (`Token MyAppName`) viene redatto.
|
|
43
|
+
# False-positive guard for the BARE scheme: what follows is NOT a credential if it has the shape
|
|
44
|
+
# of a human-written word, i.e. letters only in ONE of three forms — `lowercase`, `Capitalized`,
|
|
45
|
+
# `UPPERCASE` — and shorter than 20 characters. It spares "the bearer of bad news",
|
|
46
|
+
# "invalid bearer credentials", "Basic HTTP authentication", "Bearer Token expired".
|
|
47
|
+
# Everything else is a credential: entropy (digits, `_`, `-`, `.`), mixed inner capitals
|
|
48
|
+
# (`dTpw`, base64) or length ≥ 20 (no word is that long, while an opaque lowercase-only token is).
|
|
49
|
+
# Accepted trade-off: a CamelCase word right after the scheme (`Token MyAppName`) gets redacted.
|
|
51
50
|
PROSE_WORD = /(?:[a-z]{1,19}|[A-Z][a-z]{0,18}|[A-Z]{1,19})/
|
|
52
51
|
NOT_PROSE = %r{(?!\[FILTERED\])(?!#{PROSE_WORD}(?![A-Za-z0-9\-._~+/=]))}
|
|
53
52
|
|
|
54
|
-
#
|
|
55
|
-
#
|
|
56
|
-
#
|
|
57
|
-
#
|
|
58
|
-
#
|
|
53
|
+
# Value of a header pasted into text: a sequence of non-delimiter pieces, where a quoted string
|
|
54
|
+
# counts as ONE piece. NOT `\S+`, which would eat the `)` of "(Authorization: Bearer <token>)";
|
|
55
|
+
# but not a plain class either, which would stop at the first quote leaving in clear the secret
|
|
56
|
+
# of `Token token="<token>"` (RFC 7235 syntax and Rails' `authenticate_with_http_token`) and of a
|
|
57
|
+
# JSON fragment pasted into the message.
|
|
59
58
|
QUOTED = /"[^"\n]*"|'[^'\n]*'/
|
|
60
59
|
HEADER_VALUE = /(?:#{QUOTED}|[^\s,;)\]}"'<>])+/
|
|
61
60
|
|
|
62
|
-
# `Authorization:`/`authorization=` (
|
|
63
|
-
#
|
|
64
|
-
# over-redaction
|
|
65
|
-
#
|
|
66
|
-
#
|
|
61
|
+
# `Authorization:`/`authorization=` (also `Proxy-Authorization`, also `authorization_header`)
|
|
62
|
+
# followed by the value: here the key is explicit, so it is redacted without prose heuristics —
|
|
63
|
+
# intentional over-redaction, consistent with the DENYLIST. `[ \t]` not `\s`: never past newline.
|
|
64
|
+
# The lookahead covers the OPTIONAL scheme: without it, on already redacted text the engine would
|
|
65
|
+
# backtrack onto the "no scheme" branch and redact the word `Bearer` itself (not idempotent).
|
|
67
66
|
AUTH_HEADER = /((?i:(?:proxy[_-]?)?authorization)[\w-]*["']?[ \t]*[:=][ \t]*)(?!(?:#{AUTH_SCHEME}[ \t]+)?\[FILTERED\])(#{AUTH_SCHEME}[ \t]+)?#{HEADER_VALUE}/
|
|
68
67
|
|
|
69
|
-
#
|
|
70
|
-
#
|
|
68
|
+
# Bare scheme (`Bearer <token>` without the header name) followed by something shaped like a
|
|
69
|
+
# credential: here the prose guard is needed, the word "bearer" shows up in messages.
|
|
71
70
|
AUTH_CREDENTIAL = /(\b#{AUTH_SCHEME}[ \t]+)#{NOT_PROSE}#{CREDENTIAL_TOKEN}/
|
|
72
71
|
|
|
73
72
|
STRING_LITERAL = /'(?:[^']|'')*'/
|
|
@@ -77,7 +76,7 @@ module CloseYourIt
|
|
|
77
76
|
@configuration = configuration
|
|
78
77
|
end
|
|
79
78
|
|
|
80
|
-
#
|
|
79
|
+
# Recursively filters Hash/Array replacing the values of sensitive keys.
|
|
81
80
|
def filter_params(value)
|
|
82
81
|
case value
|
|
83
82
|
when Hash
|
|
@@ -91,15 +90,15 @@ module CloseYourIt
|
|
|
91
90
|
end
|
|
92
91
|
end
|
|
93
92
|
|
|
94
|
-
#
|
|
93
|
+
# Masks the (string/numeric) literals in SQL, preserving the structure.
|
|
95
94
|
def obfuscate_sql(sql)
|
|
96
95
|
return sql if sql.nil? || !@configuration.obfuscate_sql
|
|
97
96
|
|
|
98
97
|
sql.to_s.gsub(STRING_LITERAL, "?").gsub(NUMERIC_LITERAL, "?")
|
|
99
98
|
end
|
|
100
99
|
|
|
101
|
-
#
|
|
102
|
-
#
|
|
100
|
+
# Built-in rule (authentication schemes) BEFORE the user patterns: the two protections add up,
|
|
101
|
+
# the first cannot be disabled — privacy-by-default (CYRB-23).
|
|
103
102
|
def scrub_message(message)
|
|
104
103
|
return message if message.nil?
|
|
105
104
|
|
|
@@ -112,7 +111,7 @@ module CloseYourIt
|
|
|
112
111
|
end
|
|
113
112
|
end
|
|
114
113
|
|
|
115
|
-
#
|
|
114
|
+
# Value of a single bind/argument: redacted when the name (column/parameter) is sensitive.
|
|
116
115
|
def filter_value(key, value)
|
|
117
116
|
sensitive_key?(key) ? FILTERED : value
|
|
118
117
|
end
|
|
@@ -2,8 +2,8 @@
|
|
|
2
2
|
|
|
3
3
|
module CloseYourIt
|
|
4
4
|
module Sidekiq
|
|
5
|
-
#
|
|
6
|
-
# `call(exception, context, config)`
|
|
5
|
+
# Sidekiq error handler (registered by the railtie only when Sidekiq is present). Sidekiq calls
|
|
6
|
+
# `call(exception, context, config)` and does NOT re-raise → here we just capture.
|
|
7
7
|
class ErrorHandler
|
|
8
8
|
def call(exception, context, _config = nil)
|
|
9
9
|
apply_job_scope(context)
|
|
@@ -4,13 +4,12 @@ require_relative "../subscribers/job_performance"
|
|
|
4
4
|
|
|
5
5
|
module CloseYourIt
|
|
6
6
|
module Sidekiq
|
|
7
|
-
#
|
|
8
|
-
#
|
|
9
|
-
#
|
|
10
|
-
#
|
|
11
|
-
#
|
|
12
|
-
#
|
|
13
|
-
# decisione vive in #record).
|
|
7
|
+
# Sidekiq server middleware (registered by the railtie only when Sidekiq is present) that measures
|
|
8
|
+
# the job's execution duration and queue wait, then delegates emitting over-threshold metrics to
|
|
9
|
+
# Subscribers::JobPerformance. It does not alter the job: it times around `yield` and re-raises
|
|
10
|
+
# any error unchanged (error capture belongs to the ErrorHandler). Measurement happens in `ensure`,
|
|
11
|
+
# so it is taken for jobs that raise too; telemetry is isolated (an error in our code never
|
|
12
|
+
# disturbs the host job). Effectively a no-op when `monitor_jobs` is OFF (the decision lives in #record).
|
|
14
13
|
class JobMetricsMiddleware
|
|
15
14
|
def initialize(subscriber = nil)
|
|
16
15
|
@subscriber = subscriber
|
|
@@ -18,8 +17,8 @@ module CloseYourIt
|
|
|
18
17
|
|
|
19
18
|
def call(_worker, job, queue)
|
|
20
19
|
started = Process.clock_gettime(Process::CLOCK_MONOTONIC)
|
|
21
|
-
#
|
|
22
|
-
#
|
|
20
|
+
# The queue wait is known at the START (now - enqueued_at); Sidekiq stores enqueued_at as epoch
|
|
21
|
+
# seconds. Computed before the yield so it does not include the job duration.
|
|
23
22
|
latency = Subscribers::JobPerformance.latency_ms(job["enqueued_at"], now: Time.now.utc)
|
|
24
23
|
yield
|
|
25
24
|
ensure
|
|
@@ -47,8 +46,8 @@ module CloseYourIt
|
|
|
47
46
|
@subscriber ||= Subscribers::JobPerformance.new
|
|
48
47
|
end
|
|
49
48
|
|
|
50
|
-
#
|
|
51
|
-
#
|
|
49
|
+
# 1-based execution number. Sidekiq does NOT set `retry_count` on the first run (nil), sets it to 0
|
|
50
|
+
# on the first retry, 1 on the second, ... → attempt = retry_count + 2 when present, 1 on the first run.
|
|
52
51
|
def attempt(job)
|
|
53
52
|
count = job["retry_count"]
|
|
54
53
|
count.nil? ? 1 : count + 2
|
data/lib/closeyourit/stats.rb
CHANGED
|
@@ -3,11 +3,11 @@
|
|
|
3
3
|
require "concurrent"
|
|
4
4
|
|
|
5
5
|
module CloseYourIt
|
|
6
|
-
#
|
|
7
|
-
#
|
|
8
|
-
#
|
|
9
|
-
#
|
|
10
|
-
#
|
|
6
|
+
# Thread-safe client diagnostic counters: how many events were enqueued, dropped (full queue /
|
|
7
|
+
# before_send / sampling), sent successfully, failed (network or non-2xx status) and, among the
|
|
8
|
+
# failed, how many by network timeout. They make the silent failures of the fire-and-forget
|
|
9
|
+
# transport visible. `timeout` is a sub-count of `failed` (a timeout is still a send failure):
|
|
10
|
+
# we keep them apart to isolate connectivity problems from non-2xx.
|
|
11
11
|
#
|
|
12
12
|
# CloseYourIt.stats.to_h # => { enqueued: 12, dropped: 0, sent: 11, failed: 1, timeout: 1 }
|
|
13
13
|
class Stats
|
|
@@ -31,9 +31,9 @@ module CloseYourIt
|
|
|
31
31
|
@counters.transform_values(&:value)
|
|
32
32
|
end
|
|
33
33
|
|
|
34
|
-
#
|
|
35
|
-
# `to_h`,
|
|
36
|
-
#
|
|
34
|
+
# Thread-safe snapshot of the counters (each value read atomically). It is the same Hash as
|
|
35
|
+
# `to_h`, with an explicit name for the "read the local diagnostics" use case (CYRB-12): a pure
|
|
36
|
+
# in-memory read, it never sends telemetry.
|
|
37
37
|
alias_method :snapshot, :to_h
|
|
38
38
|
|
|
39
39
|
def reset!
|
|
@@ -5,23 +5,23 @@ require_relative "../events/job_metric_event"
|
|
|
5
5
|
|
|
6
6
|
module CloseYourIt
|
|
7
7
|
module Subscribers
|
|
8
|
-
#
|
|
9
|
-
# Sidekiq —
|
|
10
|
-
#
|
|
11
|
-
#
|
|
12
|
-
# ActiveSupport::Notifications
|
|
13
|
-
#
|
|
8
|
+
# Measures execution duration and queue wait (queue latency) of background jobs — ActiveJob and
|
|
9
|
+
# Sidekiq — and, beyond the configured thresholds, emits performance_issue metrics (subtypes
|
|
10
|
+
# `slow_job` and `job_queue_latency`). PURE logic WITHOUT shared state: everything comes in as
|
|
11
|
+
# parameters, so concurrent jobs on the same thread/process do not contaminate each other. The
|
|
12
|
+
# wiring to ActiveSupport::Notifications and the Sidekiq middleware lives elsewhere (Railtie /
|
|
13
|
+
# JobMetricsMiddleware). Honors the `monitor_jobs` master switch, the thresholds and `jobs_sample_rate`.
|
|
14
14
|
class JobPerformance
|
|
15
15
|
def initialize(configuration = nil)
|
|
16
16
|
@configuration = configuration
|
|
17
17
|
end
|
|
18
18
|
|
|
19
|
-
#
|
|
20
|
-
#
|
|
21
|
-
# fire-and-forget. `duration_ms`
|
|
22
|
-
#
|
|
23
|
-
#
|
|
24
|
-
#
|
|
19
|
+
# Single decision point: from the measured values (duration and/or wait) it builds 0..2 metrics,
|
|
20
|
+
# applies the thresholds (strict `>`, so X makes no noise and X+1 does) and sampling, then sends
|
|
21
|
+
# them fire-and-forget. `duration_ms` and `queue_latency_ms` are optional: ActiveJob provides them
|
|
22
|
+
# from two separate hooks (perform_start → wait, perform → duration), Sidekiq both in one call.
|
|
23
|
+
# Sampling is applied ONLY to candidates already beyond the threshold (normal jobs neither
|
|
24
|
+
# consume nor generate anything).
|
|
25
25
|
def record(job_class:, queue: nil, adapter: nil, duration_ms: nil, queue_latency_ms: nil,
|
|
26
26
|
attempt: nil, trace_id: nil)
|
|
27
27
|
config = configuration
|
|
@@ -38,19 +38,20 @@ module CloseYourIt
|
|
|
38
38
|
nil
|
|
39
39
|
end
|
|
40
40
|
|
|
41
|
-
#
|
|
41
|
+
# `perform_start.active_job` hook: the queue wait is known as soon as the job starts (now - the
|
|
42
|
+
# moment it was due to run).
|
|
42
43
|
def active_job_started(job, now: Time.now.utc)
|
|
43
44
|
record(
|
|
44
45
|
job_class: job.class.name,
|
|
45
46
|
queue: (job.queue_name if job.respond_to?(:queue_name)),
|
|
46
47
|
adapter: "active_job",
|
|
47
|
-
queue_latency_ms: self.class.latency_ms(
|
|
48
|
+
queue_latency_ms: self.class.latency_ms(due_at(job), now: now),
|
|
48
49
|
attempt: (job.executions if job.respond_to?(:executions)),
|
|
49
50
|
trace_id: (job.job_id if job.respond_to?(:job_id))
|
|
50
51
|
)
|
|
51
52
|
end
|
|
52
53
|
|
|
53
|
-
#
|
|
54
|
+
# `perform.active_job` hook: at the end of execution the duration is `event.duration` (ms).
|
|
54
55
|
def active_job_performed(job, duration_ms)
|
|
55
56
|
record(
|
|
56
57
|
job_class: job.class.name,
|
|
@@ -62,10 +63,10 @@ module CloseYourIt
|
|
|
62
63
|
)
|
|
63
64
|
end
|
|
64
65
|
|
|
65
|
-
#
|
|
66
|
-
#
|
|
67
|
-
#
|
|
68
|
-
#
|
|
66
|
+
# Normalizes `enqueued_at` (Time, Numeric epoch in seconds like Sidekiq, or ISO8601 String) into
|
|
67
|
+
# a wait (ms) relative to `now`. nil or unparsable → nil: no wait metric (adapter that does not
|
|
68
|
+
# populate the enqueue instant). Clamped to 0 when negative (clock skew, enqueue "in the future"):
|
|
69
|
+
# the ingest schema requires `duration_ms >= 0`.
|
|
69
70
|
def self.latency_ms(enqueued_at, now:)
|
|
70
71
|
started = to_time(enqueued_at)
|
|
71
72
|
return nil if started.nil?
|
|
@@ -94,6 +95,16 @@ module CloseYourIt
|
|
|
94
95
|
job.enqueued_at if job.respond_to?(:enqueued_at)
|
|
95
96
|
end
|
|
96
97
|
|
|
98
|
+
# A job delayed on purpose (set(wait:), retry_on backoff) is due at scheduled_at, not when it was
|
|
99
|
+
# enqueued: the planned delay is not queue wait (CYRB-27). No enqueued_at still means no metric.
|
|
100
|
+
def due_at(job)
|
|
101
|
+
enqueued = self.class.to_time(enqueued_at(job))
|
|
102
|
+
scheduled = self.class.to_time(job.scheduled_at) if job.respond_to?(:scheduled_at)
|
|
103
|
+
return enqueued if enqueued.nil? || scheduled.nil?
|
|
104
|
+
|
|
105
|
+
[ enqueued, scheduled ].max
|
|
106
|
+
end
|
|
107
|
+
|
|
97
108
|
def configuration
|
|
98
109
|
@configuration || CloseYourIt.configuration
|
|
99
110
|
end
|