tina4ruby 3.13.161 → 3.13.162
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/CHANGELOG.md +6 -0
- data/lib/tina4/queue_backends/kafka_backend.rb +40 -20
- data/lib/tina4/shutdown.rb +36 -7
- data/lib/tina4/version.rb +1 -1
- metadata +2 -2
checksums.yaml
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
---
|
|
2
2
|
SHA256:
|
|
3
|
-
metadata.gz:
|
|
4
|
-
data.tar.gz:
|
|
3
|
+
metadata.gz: 15151f90c7170c25e2174fdf43bbf5c847c65478b082dd5bdbda8049a5b57b97
|
|
4
|
+
data.tar.gz: a040ce2ecad0cfe976a1eb579c001e07c4f9f27244f8c1e35591ba1b536ea8cc
|
|
5
5
|
SHA512:
|
|
6
|
-
metadata.gz:
|
|
7
|
-
data.tar.gz:
|
|
6
|
+
metadata.gz: e1a7d0ff48f5310c2ee54a8d33c78bdfdcbdb1480a7633136cbde22cdea58c2e8b4f27845d1a21a14c79e4f4789c8b9fc009ec0bd61026a639ee0842598279c4
|
|
7
|
+
data.tar.gz: 1bead80a66a17fe802aac1efdd010695e884966c0b6848bfe18d6892899bd6f48e0854303afb67f303532649302201eb85ffcee7308b2cef304e2cd03f9ae2e4
|
data/CHANGELOG.md
CHANGED
|
@@ -6,6 +6,12 @@ number means the same thing everywhere.
|
|
|
6
6
|
**The authoritative release notes for every shipped version live in the documentation:**
|
|
7
7
|
https://tina4.com/ruby/36-releases
|
|
8
8
|
|
|
9
|
+
## 3.13.162 - 2026-11-10
|
|
10
|
+
|
|
11
|
+
### Fixed
|
|
12
|
+
|
|
13
|
+
- Kafka tombstone (null-value) records no longer crash the consumer (tina4-python#303 twin). A tombstone — a record whose value is `null` (Debezium deletes, log-compacted delete markers) — reached `Tina4.parse_json(msg.payload)` on the rdkafka path, and `JSON.parse(nil)` raised `TypeError`, which `rescue Rdkafka::RdkafkaError` did not catch, so it escaped and killed the consumer; because the tombstone sits at the committed offset it was a poison pill that re-crashed on every restart. `dequeue` and the `dead_letters` reader now skip a `nil`-payload record and continue, so a following real record is still delivered. No mocks — driven against a real librdkafka producer/broker (#167).
|
|
14
|
+
|
|
9
15
|
## 3.13.161 - 2026-11-09
|
|
10
16
|
|
|
11
17
|
### Fixed
|
|
@@ -161,26 +161,8 @@ module Tina4
|
|
|
161
161
|
end
|
|
162
162
|
return nil unless msg
|
|
163
163
|
|
|
164
|
-
|
|
165
|
-
|
|
166
|
-
|
|
167
|
-
# A record carrying a "payload" key is a Tina4 envelope: keep its own id
|
|
168
|
-
# (falling back to the coordinate) and carry attempts/error back so a
|
|
169
|
-
# redelivery does not reset the count and re-produce a poison record
|
|
170
|
-
# forever. A record WITHOUT one is externally produced (Debezium/CDC and
|
|
171
|
-
# friends, no envelope, no id -- tina4#274): its whole value IS the
|
|
172
|
-
# payload, and the coordinate becomes its id. Either way the job gets the
|
|
173
|
-
# id acknowledge() maps back to this record's offset.
|
|
174
|
-
job =
|
|
175
|
-
if data.is_a?(Hash) && data.key?("payload")
|
|
176
|
-
id = data["id"].to_s.empty? ? coordinate : data["id"]
|
|
177
|
-
Tina4::Job.new(
|
|
178
|
-
topic: topic, payload: data["payload"], id: id,
|
|
179
|
-
attempts: data["attempts"] || 0, error: data["error"]
|
|
180
|
-
)
|
|
181
|
-
else
|
|
182
|
-
Tina4::Job.new(topic: topic, payload: data, id: coordinate, attempts: 0)
|
|
183
|
-
end
|
|
164
|
+
job = job_from_message(topic, msg)
|
|
165
|
+
return nil unless job
|
|
184
166
|
|
|
185
167
|
@in_flight[job.id.to_s] = msg
|
|
186
168
|
job
|
|
@@ -188,6 +170,39 @@ module Tina4
|
|
|
188
170
|
nil
|
|
189
171
|
end
|
|
190
172
|
|
|
173
|
+
# Build a Job from a polled record, or nil for a record that carries no job.
|
|
174
|
+
#
|
|
175
|
+
# A null-value record is a tombstone / delete marker: Debezium emits these on
|
|
176
|
+
# deletes and log-compacted topics use them as delete markers, so they are
|
|
177
|
+
# normal on Kafka. rdkafka hands back a nil payload for one; return nil (skip
|
|
178
|
+
# it) so dequeue advances past it. Before this guard, JSON.parse(nil) raised a
|
|
179
|
+
# TypeError that dequeue's `rescue Rdkafka::RdkafkaError` did NOT catch, so the
|
|
180
|
+
# tombstone was a poison pill that killed the consumer on every restart
|
|
181
|
+
# (tina4-python#303, same crash in the rdkafka path).
|
|
182
|
+
#
|
|
183
|
+
# A record carrying a "payload" key is a Tina4 envelope: keep its own id
|
|
184
|
+
# (falling back to the coordinate) and carry attempts/error back so a
|
|
185
|
+
# redelivery does not reset the count and re-produce a poison record forever.
|
|
186
|
+
# A record WITHOUT one is externally produced (Debezium/CDC and friends, no
|
|
187
|
+
# envelope, no id -- tina4#274): its whole value IS the payload, and the
|
|
188
|
+
# coordinate becomes its id. Either way the job gets the id acknowledge()
|
|
189
|
+
# maps back to this record's offset.
|
|
190
|
+
def job_from_message(topic, msg)
|
|
191
|
+
return nil if msg.payload.nil?
|
|
192
|
+
|
|
193
|
+
data = Tina4.parse_json(msg.payload)
|
|
194
|
+
coordinate = "#{topic}:#{msg.partition}:#{msg.offset}"
|
|
195
|
+
if data.is_a?(Hash) && data.key?("payload")
|
|
196
|
+
id = data["id"].to_s.empty? ? coordinate : data["id"]
|
|
197
|
+
Tina4::Job.new(
|
|
198
|
+
topic: topic, payload: data["payload"], id: id,
|
|
199
|
+
attempts: data["attempts"] || 0, error: data["error"]
|
|
200
|
+
)
|
|
201
|
+
else
|
|
202
|
+
Tina4::Job.new(topic: topic, payload: data, id: coordinate, attempts: 0)
|
|
203
|
+
end
|
|
204
|
+
end
|
|
205
|
+
|
|
191
206
|
# Kafka has no queue depth: a log is a sequence of offsets, not a queue,
|
|
192
207
|
# and computing a "remaining" count means an admin round-trip per call.
|
|
193
208
|
# ADR-0022 decision 5 records 0 as the documented answer, which Python and
|
|
@@ -287,6 +302,11 @@ module Tina4
|
|
|
287
302
|
loop do
|
|
288
303
|
msg = reader.poll(500)
|
|
289
304
|
if msg
|
|
305
|
+
# Skip a tombstone / null-value record the same way dequeue does, so
|
|
306
|
+
# a null-value dead-letter never crashes the reader on JSON.parse(nil)
|
|
307
|
+
# (tina4-python#303).
|
|
308
|
+
next if msg.payload.nil?
|
|
309
|
+
|
|
290
310
|
data = Tina4.parse_json(msg.payload)
|
|
291
311
|
data["status"] = "dead"
|
|
292
312
|
out << data
|
data/lib/tina4/shutdown.rb
CHANGED
|
@@ -72,17 +72,29 @@ module Tina4
|
|
|
72
72
|
return if @shutting_down
|
|
73
73
|
|
|
74
74
|
@shutting_down = true
|
|
75
|
-
Tina4::Log.info("Shutdown signal received, stopping gracefully...")
|
|
76
75
|
|
|
77
|
-
|
|
78
|
-
|
|
79
|
-
|
|
80
|
-
|
|
81
|
-
|
|
76
|
+
# The whole teardown runs INSIDE the signal handler's thread
|
|
77
|
+
# (install_signal_handlers). An exception escaping here - a failing log
|
|
78
|
+
# sink under TINA4_LOG_STRICT, a half-closed WebSocket, a dead database
|
|
79
|
+
# driver, Log.reset - would kill that thread mid-teardown. If it dies
|
|
80
|
+
# BEFORE stop_accepting, the listening socket never closes and the
|
|
81
|
+
# process HANGS on a signal that MUST drain and exit 0; if it dies
|
|
82
|
+
# later, resources leak. So every step degrades independently
|
|
83
|
+
# (guard_teardown) and the critical ones still run even when an earlier
|
|
84
|
+
# step threw - mirrors the per-step try/catch(\Throwable) in PHP's
|
|
85
|
+
# Server::cleanup()/stop().
|
|
86
|
+
guard_teardown { Tina4::Log.info("Shutdown signal received, stopping gracefully...") }
|
|
87
|
+
guard_teardown { stop_accepting }
|
|
88
|
+
|
|
89
|
+
drained = true
|
|
90
|
+
guard_teardown { drained = wait_for_in_flight }
|
|
91
|
+
guard_teardown { release_resources }
|
|
92
|
+
|
|
93
|
+
guard_teardown { Tina4::Log.info("Shutdown complete") }
|
|
82
94
|
# Graceful shutdown owns the final call to reset() (Decision 24 /
|
|
83
95
|
# LOG-I02): flush+close owned sinks AFTER the shutdown record above,
|
|
84
96
|
# exactly once.
|
|
85
|
-
Tina4::Log.reset
|
|
97
|
+
guard_teardown { Tina4::Log.reset }
|
|
86
98
|
@shutdown_complete = true
|
|
87
99
|
return if drained
|
|
88
100
|
|
|
@@ -139,6 +151,23 @@ module Tina4
|
|
|
139
151
|
|
|
140
152
|
private
|
|
141
153
|
|
|
154
|
+
# Run one teardown step so that NO exception escapes the signal handler's
|
|
155
|
+
# thread (see initiate_shutdown). On a failure, degrade to a best-effort
|
|
156
|
+
# Tina4::Log.error; if the logger itself is what failed (a strict log sink
|
|
157
|
+
# that raises on write), fall back to $stderr so the degrade is never
|
|
158
|
+
# silent - then carry on, so the shutdown still completes and the process
|
|
159
|
+
# still exits 0. Keeps shutdown idempotent and exception-safe without
|
|
160
|
+
# changing the teardown order.
|
|
161
|
+
def guard_teardown
|
|
162
|
+
yield
|
|
163
|
+
rescue StandardError => e
|
|
164
|
+
begin
|
|
165
|
+
Tina4::Log.error("Error during graceful shutdown: #{e.class}: #{e.message}")
|
|
166
|
+
rescue StandardError
|
|
167
|
+
warn("tina4: graceful shutdown step failed: #{e.class}: #{e.message}")
|
|
168
|
+
end
|
|
169
|
+
end
|
|
170
|
+
|
|
142
171
|
# TINA4_SHUTDOWN_TIMEOUT in seconds (same name and default in all four
|
|
143
172
|
# frameworks). An unparseable or negative value warns and falls back to the
|
|
144
173
|
# default: `"abc".to_i` is 0, and a silent 0-second drain would force-close
|
data/lib/tina4/version.rb
CHANGED
metadata
CHANGED
|
@@ -1,14 +1,14 @@
|
|
|
1
1
|
--- !ruby/object:Gem::Specification
|
|
2
2
|
name: tina4ruby
|
|
3
3
|
version: !ruby/object:Gem::Version
|
|
4
|
-
version: 3.13.
|
|
4
|
+
version: 3.13.162
|
|
5
5
|
platform: ruby
|
|
6
6
|
authors:
|
|
7
7
|
- Tina4 Team
|
|
8
8
|
autorequire:
|
|
9
9
|
bindir: exe
|
|
10
10
|
cert_chain: []
|
|
11
|
-
date: 2026-11-
|
|
11
|
+
date: 2026-11-12 00:00:00.000000000 Z
|
|
12
12
|
dependencies:
|
|
13
13
|
- !ruby/object:Gem::Dependency
|
|
14
14
|
name: sqlite3
|