karafka 2.3.0 → 2.3.2
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- checksums.yaml.gz.sig +0 -0
- data/.rspec +2 -0
- data/CHANGELOG.md +15 -0
- data/Gemfile +1 -1
- data/Gemfile.lock +22 -22
- data/README.md +2 -2
- data/bin/integrations +2 -1
- data/bin/rspecs +6 -2
- data/config/locales/errors.yml +30 -8
- data/config/locales/pro_errors.yml +2 -0
- data/docker-compose.yml +1 -1
- data/lib/karafka/app.rb +14 -0
- data/lib/karafka/cli/base.rb +19 -0
- data/lib/karafka/cli/server.rb +62 -76
- data/lib/karafka/cli/swarm.rb +30 -0
- data/lib/karafka/constraints.rb +3 -3
- data/lib/karafka/contracts/config.rb +19 -0
- data/lib/karafka/errors.rb +12 -0
- data/lib/karafka/helpers/async.rb +13 -3
- data/lib/karafka/helpers/config_importer.rb +30 -0
- data/lib/karafka/instrumentation/logger_listener.rb +31 -0
- data/lib/karafka/instrumentation/notifications.rb +9 -0
- data/lib/karafka/instrumentation/vendors/datadog/logger_listener.rb +2 -0
- data/lib/karafka/instrumentation/vendors/kubernetes/base_listener.rb +72 -0
- data/lib/karafka/instrumentation/vendors/kubernetes/liveness_listener.rb +11 -40
- data/lib/karafka/instrumentation/vendors/kubernetes/swarm_liveness_listener.rb +54 -0
- data/lib/karafka/pro/active_job/job_options_contract.rb +1 -1
- data/lib/karafka/pro/base_consumer.rb +16 -0
- data/lib/karafka/pro/connection/manager.rb +6 -1
- data/lib/karafka/pro/processing/coordinator.rb +13 -3
- data/lib/karafka/pro/processing/coordinators/errors_tracker.rb +74 -0
- data/lib/karafka/pro/processing/coordinators/filters_applier.rb +107 -0
- data/lib/karafka/pro/processing/coordinators/virtual_offset_manager.rb +180 -0
- data/lib/karafka/pro/processing/strategies/aj/dlq_ftr_lrj_mom.rb +5 -7
- data/lib/karafka/pro/processing/strategies/aj/dlq_ftr_lrj_mom_vp.rb +5 -7
- data/lib/karafka/pro/processing/strategies/aj/dlq_ftr_mom.rb +8 -10
- data/lib/karafka/pro/processing/strategies/aj/dlq_ftr_mom_vp.rb +8 -16
- data/lib/karafka/pro/processing/strategies/aj/dlq_lrj_mom.rb +5 -7
- data/lib/karafka/pro/processing/strategies/aj/dlq_lrj_mom_vp.rb +5 -7
- data/lib/karafka/pro/processing/strategies/aj/dlq_mom.rb +8 -10
- data/lib/karafka/pro/processing/strategies/aj/dlq_mom_vp.rb +7 -9
- data/lib/karafka/pro/processing/strategies/dlq/default.rb +36 -10
- data/lib/karafka/pro/processing/strategies/dlq/ftr.rb +3 -7
- data/lib/karafka/pro/processing/strategies/dlq/ftr_lrj.rb +4 -8
- data/lib/karafka/pro/processing/strategies/dlq/ftr_lrj_mom.rb +6 -9
- data/lib/karafka/pro/processing/strategies/dlq/ftr_mom.rb +5 -15
- data/lib/karafka/pro/processing/strategies/dlq/lrj.rb +4 -8
- data/lib/karafka/pro/processing/strategies/dlq/lrj_mom.rb +6 -9
- data/lib/karafka/pro/processing/strategies/dlq/mom.rb +10 -20
- data/lib/karafka/pro/processing/strategies/vp/default.rb +7 -0
- data/lib/karafka/pro/routing/features/dead_letter_queue/contracts/topic.rb +6 -0
- data/lib/karafka/pro/routing/features/dead_letter_queue/topic.rb +39 -0
- data/lib/karafka/pro/swarm/liveness_listener.rb +171 -0
- data/lib/karafka/process.rb +27 -1
- data/lib/karafka/routing/features/dead_letter_queue/config.rb +2 -0
- data/lib/karafka/routing/subscription_group.rb +31 -9
- data/lib/karafka/runner.rb +4 -0
- data/lib/karafka/server.rb +13 -16
- data/lib/karafka/setup/config.rb +41 -2
- data/lib/karafka/status.rb +4 -2
- data/lib/karafka/swarm/liveness_listener.rb +55 -0
- data/lib/karafka/swarm/manager.rb +217 -0
- data/lib/karafka/swarm/node.rb +179 -0
- data/lib/karafka/swarm/pidfd.rb +131 -0
- data/lib/karafka/swarm/supervisor.rb +184 -0
- data/lib/karafka/swarm.rb +27 -0
- data/lib/karafka/templates/karafka.rb.erb +0 -2
- data/lib/karafka/version.rb +1 -1
- data/lib/karafka.rb +1 -1
- data.tar.gz.sig +0 -0
- metadata +17 -4
- metadata.gz.sig +0 -0
- data/lib/karafka/pro/processing/filters_applier.rb +0 -105
- data/lib/karafka/pro/processing/virtual_offset_manager.rb +0 -177
@@ -226,6 +226,34 @@ module Karafka
|
|
226
226
|
MSG
|
227
227
|
end
|
228
228
|
|
229
|
+
# @param event [Karafka::Core::Monitoring::Event] event details including payload
|
230
|
+
def on_swarm_manager_stopping(event)
|
231
|
+
node = event[:node]
|
232
|
+
error "Swarm manager detected unhealthy node #{node.pid}. Sending TERM signal..."
|
233
|
+
end
|
234
|
+
|
235
|
+
# @param event [Karafka::Core::Monitoring::Event] event details including payload
|
236
|
+
def on_swarm_manager_terminating(event)
|
237
|
+
node = event[:node]
|
238
|
+
error "Swarm manager detected unresponsive node #{node.pid}. Sending KILL signal..."
|
239
|
+
end
|
240
|
+
|
241
|
+
# @param event [Karafka::Core::Monitoring::Event] event details including payload
|
242
|
+
def on_swarm_manager_before_fork(event)
|
243
|
+
debug "Swarm manager starting node with id: #{event[:node].id}"
|
244
|
+
end
|
245
|
+
|
246
|
+
# @param _event [Karafka::Core::Monitoring::Event] event details including payload
|
247
|
+
def on_swarm_node_after_fork(_event)
|
248
|
+
info "Swarm node #{::Process.pid} forked from #{::Process.ppid}"
|
249
|
+
end
|
250
|
+
|
251
|
+
# @param event [Karafka::Core::Monitoring::Event] event details including payload
|
252
|
+
def on_swarm_manager_control(event)
|
253
|
+
pids = event[:caller].nodes.map(&:pid).join(', ')
|
254
|
+
debug "Swarm manager checking nodes: #{pids}"
|
255
|
+
end
|
256
|
+
|
229
257
|
# There are many types of errors that can occur in many places, but we provide a single
|
230
258
|
# handler for all of them to simplify error instrumentation.
|
231
259
|
# @param event [Karafka::Core::Monitoring::Event] event details including payload
|
@@ -259,6 +287,9 @@ module Karafka
|
|
259
287
|
when 'connection.listener.fetch_loop.error'
|
260
288
|
error "Listener fetch loop error: #{error}"
|
261
289
|
error details
|
290
|
+
when 'swarm.supervisor.error'
|
291
|
+
fatal "Swarm supervisor crashed due to an error: #{error}"
|
292
|
+
fatal details
|
262
293
|
when 'runner.call.error'
|
263
294
|
fatal "Runner crashed due to an error: #{error}"
|
264
295
|
fatal details
|
@@ -22,6 +22,8 @@ module Karafka
|
|
22
22
|
|
23
23
|
app.initializing
|
24
24
|
app.initialized
|
25
|
+
app.before_warmup
|
26
|
+
app.supervising
|
25
27
|
app.running
|
26
28
|
app.quieting
|
27
29
|
app.quiet
|
@@ -73,6 +75,13 @@ module Karafka
|
|
73
75
|
|
74
76
|
statistics.emitted
|
75
77
|
|
78
|
+
swarm.node.after_fork
|
79
|
+
swarm.manager.before_fork
|
80
|
+
swarm.manager.after_fork
|
81
|
+
swarm.manager.control
|
82
|
+
swarm.manager.stopping
|
83
|
+
swarm.manager.terminating
|
84
|
+
|
76
85
|
worker.process
|
77
86
|
worker.processed
|
78
87
|
worker.completed
|
@@ -129,6 +129,8 @@ module Karafka
|
|
129
129
|
fatal "Runner crashed due to an error: #{error}"
|
130
130
|
when 'app.stopping.error'
|
131
131
|
error 'Forceful Karafka server stop'
|
132
|
+
when 'swarm.supervisor.error'
|
133
|
+
fatal "Swarm supervisor crashed due to an error: #{error}"
|
132
134
|
when 'librdkafka.error'
|
133
135
|
error "librdkafka internal error occurred: #{error}"
|
134
136
|
# Those will only occur when retries in the client fail and when they did not stop
|
@@ -0,0 +1,72 @@
|
|
1
|
+
# frozen_string_literal: true
|
2
|
+
|
3
|
+
require 'socket'
|
4
|
+
|
5
|
+
module Karafka
|
6
|
+
module Instrumentation
|
7
|
+
module Vendors
|
8
|
+
# Namespace for instrumentation related with Kubernetes
|
9
|
+
module Kubernetes
|
10
|
+
# Base Kubernetes Listener providing basic HTTP server capabilities to respond with health
|
11
|
+
class BaseListener
|
12
|
+
include ::Karafka::Core::Helpers::Time
|
13
|
+
|
14
|
+
# All good with Karafka
|
15
|
+
OK_CODE = '204 No Content'
|
16
|
+
|
17
|
+
# Some timeouts, fail
|
18
|
+
FAIL_CODE = '500 Internal Server Error'
|
19
|
+
|
20
|
+
private_constant :OK_CODE, :FAIL_CODE
|
21
|
+
|
22
|
+
# @param hostname [String, nil] hostname or nil to bind on all
|
23
|
+
# @param port [Integer] TCP port on which we want to run our HTTP status server
|
24
|
+
def initialize(
|
25
|
+
hostname: nil,
|
26
|
+
port: 3000
|
27
|
+
)
|
28
|
+
@hostname = hostname
|
29
|
+
@port = port
|
30
|
+
end
|
31
|
+
|
32
|
+
private
|
33
|
+
|
34
|
+
# @return [Boolean] true if all good, false if we should tell k8s to kill this process
|
35
|
+
def healthy?
|
36
|
+
raise NotImplementedError, 'Implement in a subclass'
|
37
|
+
end
|
38
|
+
|
39
|
+
# Responds to a HTTP request with the process liveness status
|
40
|
+
def respond
|
41
|
+
client = @server.accept
|
42
|
+
client.gets
|
43
|
+
client.print "HTTP/1.1 #{healthy? ? OK_CODE : FAIL_CODE}\r\n"
|
44
|
+
client.print "Content-Type: text/plain\r\n"
|
45
|
+
client.print "\r\n"
|
46
|
+
client.close
|
47
|
+
|
48
|
+
true
|
49
|
+
rescue Errno::ECONNRESET, Errno::EPIPE, IOError
|
50
|
+
!@server.closed?
|
51
|
+
end
|
52
|
+
|
53
|
+
# Starts background thread with micro-http monitoring
|
54
|
+
def start
|
55
|
+
@server = TCPServer.new(*[@hostname, @port].compact)
|
56
|
+
|
57
|
+
Thread.new do
|
58
|
+
loop do
|
59
|
+
break unless respond
|
60
|
+
end
|
61
|
+
end
|
62
|
+
end
|
63
|
+
|
64
|
+
# Stops the server
|
65
|
+
def stop
|
66
|
+
@server.close
|
67
|
+
end
|
68
|
+
end
|
69
|
+
end
|
70
|
+
end
|
71
|
+
end
|
72
|
+
end
|
@@ -1,6 +1,6 @@
|
|
1
1
|
# frozen_string_literal: true
|
2
2
|
|
3
|
-
require '
|
3
|
+
require 'karafka/instrumentation/vendors/kubernetes/base_listener'
|
4
4
|
|
5
5
|
module Karafka
|
6
6
|
module Instrumentation
|
@@ -23,17 +23,9 @@ module Karafka
|
|
23
23
|
#
|
24
24
|
# @note In case of usage within an embedding with Puma, you need to select different port
|
25
25
|
# then the one used by Puma itself.
|
26
|
-
|
27
|
-
|
28
|
-
|
29
|
-
# All good with Karafka
|
30
|
-
OK_CODE = '204 No Content'
|
31
|
-
|
32
|
-
# Some timeouts, fail
|
33
|
-
FAIL_CODE = '500 Internal Server Error'
|
34
|
-
|
35
|
-
private_constant :OK_CODE, :FAIL_CODE
|
36
|
-
|
26
|
+
#
|
27
|
+
# @note Please use `Kubernetes::SwarmLivenessListener` when operating in the swarm mode
|
28
|
+
class LivenessListener < BaseListener
|
37
29
|
# @param hostname [String, nil] hostname or nil to bind on all
|
38
30
|
# @param port [Integer] TCP port on which we want to run our HTTP status server
|
39
31
|
# @param consuming_ttl [Integer] time in ms after which we consider consumption hanging.
|
@@ -48,30 +40,23 @@ module Karafka
|
|
48
40
|
consuming_ttl: 5 * 60 * 1_000,
|
49
41
|
polling_ttl: 5 * 60 * 1_000
|
50
42
|
)
|
51
|
-
@hostname = hostname
|
52
|
-
@port = port
|
53
43
|
@polling_ttl = polling_ttl
|
54
44
|
@consuming_ttl = consuming_ttl
|
55
45
|
@mutex = Mutex.new
|
56
46
|
@pollings = {}
|
57
47
|
@consumptions = {}
|
48
|
+
super(hostname: hostname, port: port)
|
58
49
|
end
|
59
50
|
|
60
51
|
# @param _event [Karafka::Core::Monitoring::Event]
|
61
52
|
def on_app_running(_event)
|
62
|
-
|
63
|
-
|
64
|
-
Thread.new do
|
65
|
-
loop do
|
66
|
-
break unless respond
|
67
|
-
end
|
68
|
-
end
|
53
|
+
start
|
69
54
|
end
|
70
55
|
|
71
56
|
# Stop the http server when we stop the process
|
72
57
|
# @param _event [Karafka::Core::Monitoring::Event]
|
73
58
|
def on_app_stopped(_event)
|
74
|
-
|
59
|
+
stop
|
75
60
|
end
|
76
61
|
|
77
62
|
# Tick on each fetch
|
@@ -148,29 +133,15 @@ module Karafka
|
|
148
133
|
end
|
149
134
|
end
|
150
135
|
|
151
|
-
# Responds to a HTTP request with the process liveness status
|
152
|
-
def respond
|
153
|
-
client = @server.accept
|
154
|
-
client.gets
|
155
|
-
client.print "HTTP/1.1 #{status}\r\n"
|
156
|
-
client.print "Content-Type: text/plain\r\n"
|
157
|
-
client.print "\r\n"
|
158
|
-
client.close
|
159
|
-
|
160
|
-
true
|
161
|
-
rescue Errno::ECONNRESET, Errno::EPIPE, IOError
|
162
|
-
!@server.closed?
|
163
|
-
end
|
164
|
-
|
165
136
|
# Did we exceed any of the ttls
|
166
137
|
# @return [String] 204 string if ok, 500 otherwise
|
167
|
-
def
|
138
|
+
def healthy?
|
168
139
|
time = monotonic_now
|
169
140
|
|
170
|
-
return
|
171
|
-
return
|
141
|
+
return false if @pollings.values.any? { |tick| (time - tick) > @polling_ttl }
|
142
|
+
return false if @consumptions.values.any? { |tick| (time - tick) > @consuming_ttl }
|
172
143
|
|
173
|
-
|
144
|
+
true
|
174
145
|
end
|
175
146
|
end
|
176
147
|
end
|
@@ -0,0 +1,54 @@
|
|
1
|
+
# frozen_string_literal: true
|
2
|
+
|
3
|
+
require 'karafka/instrumentation/vendors/kubernetes/base_listener'
|
4
|
+
|
5
|
+
module Karafka
|
6
|
+
module Instrumentation
|
7
|
+
module Vendors
|
8
|
+
module Kubernetes
|
9
|
+
# Kubernetes HTTP listener designed to operate with Karafka running in the swarm mode
|
10
|
+
# In the Swarm mode we supervise only the supervisor as other nodes are suppose to be
|
11
|
+
# managed by the swarm supervisor
|
12
|
+
class SwarmLivenessListener < BaseListener
|
13
|
+
# @param hostname [String, nil] hostname or nil to bind on all
|
14
|
+
# @param port [Integer] TCP port on which we want to run our HTTP status server
|
15
|
+
# @param controlling_ttl [Integer] time in ms after which we consider the supervising
|
16
|
+
# thread dead because it is not controlling nodes. When configuring this, please take
|
17
|
+
# into consideration, that during shutdown of the swarm, there is no controlling
|
18
|
+
# happening.
|
19
|
+
def initialize(
|
20
|
+
hostname: nil,
|
21
|
+
port: 3000,
|
22
|
+
controlling_ttl: 60 * 1_000
|
23
|
+
)
|
24
|
+
@hostname = hostname
|
25
|
+
@port = port
|
26
|
+
@controlling_ttl = controlling_ttl
|
27
|
+
@controlling = monotonic_now
|
28
|
+
super(port: port, hostname: hostname)
|
29
|
+
end
|
30
|
+
|
31
|
+
# Starts reporting in the supervisor only when it runs
|
32
|
+
# @param _event [Karafka::Core::Monitoring::Event]
|
33
|
+
def on_app_supervising(_event)
|
34
|
+
start
|
35
|
+
end
|
36
|
+
|
37
|
+
# Tick on each control
|
38
|
+
# @param _event [Karafka::Core::Monitoring::Event]
|
39
|
+
def on_swarm_manager_control(_event)
|
40
|
+
@controlling = monotonic_now
|
41
|
+
end
|
42
|
+
|
43
|
+
private
|
44
|
+
|
45
|
+
# Did we exceed any of the ttls
|
46
|
+
# @return [String] 204 string if ok, 500 otherwise
|
47
|
+
def healthy?
|
48
|
+
(monotonic_now - @controlling) < @controlling_ttl
|
49
|
+
end
|
50
|
+
end
|
51
|
+
end
|
52
|
+
end
|
53
|
+
end
|
54
|
+
end
|
@@ -26,7 +26,7 @@ module Karafka
|
|
26
26
|
end
|
27
27
|
|
28
28
|
optional(:partitioner) { |val| val.respond_to?(:call) }
|
29
|
-
optional(:partition_key_type) { |val| %i[key partition_key].include?(val) }
|
29
|
+
optional(:partition_key_type) { |val| %i[key partition_key partition].include?(val) }
|
30
30
|
optional(:dispatch_method) do |val|
|
31
31
|
%i[
|
32
32
|
produce_async
|
@@ -42,6 +42,22 @@ module Karafka
|
|
42
42
|
|
43
43
|
# By default we do nothing when ticking
|
44
44
|
def tick; end
|
45
|
+
|
46
|
+
# @return [Karafka::Pro::Processing::Coordinators::ErrorsTracker] tracker for errors that
|
47
|
+
# occurred during processing until another successful processing
|
48
|
+
#
|
49
|
+
# @note This will always contain **only** details of errors that occurred during `#consume`
|
50
|
+
# because only those are retryable.
|
51
|
+
#
|
52
|
+
# @note This may contain more than one error because:
|
53
|
+
# - this can collect various errors that might have happened during virtual partitions
|
54
|
+
# execution
|
55
|
+
# - errors can pile up during retries and until a clean run, they will be collected with
|
56
|
+
# a limit of last 100. We do not store more because a consumer with an endless error loop
|
57
|
+
# would cause memory leaks without such a limit.
|
58
|
+
def errors_tracker
|
59
|
+
coordinator.errors_tracker
|
60
|
+
end
|
45
61
|
end
|
46
62
|
end
|
47
63
|
end
|
@@ -114,10 +114,15 @@ module Karafka
|
|
114
114
|
once(:quiet!) { active_listeners.each(&:quiet!) }
|
115
115
|
|
116
116
|
# If we are in the process of moving to quiet state, we need to check it.
|
117
|
-
if Karafka::App.quieting?
|
117
|
+
if Karafka::App.quieting?
|
118
|
+
# If we are quieting but not all active listeners are quiet we need to wait for all of
|
119
|
+
# them to reach the quiet state
|
120
|
+
return unless active_listeners.all?(&:quiet?)
|
121
|
+
|
118
122
|
once(:quieted!) { Karafka::App.quieted! }
|
119
123
|
end
|
120
124
|
|
125
|
+
# Do nothing if we moved to quiet state and want to be in it
|
121
126
|
return if Karafka::App.quiet?
|
122
127
|
|
123
128
|
# Since separate subscription groups are subscribed to different topics, there is no risk
|
@@ -21,13 +21,14 @@ module Karafka
|
|
21
21
|
|
22
22
|
def_delegators :@collapser, :collapsed?, :collapse_until!
|
23
23
|
|
24
|
-
attr_reader :filter, :virtual_offset_manager, :shared_mutex
|
24
|
+
attr_reader :filter, :virtual_offset_manager, :shared_mutex, :errors_tracker
|
25
25
|
|
26
26
|
# @param args [Object] anything the base coordinator accepts
|
27
27
|
def initialize(*args)
|
28
28
|
super
|
29
29
|
|
30
30
|
@executed = []
|
31
|
+
@errors_tracker = Coordinators::ErrorsTracker.new
|
31
32
|
@flow_mutex = Mutex.new
|
32
33
|
# Lock for user code synchronization
|
33
34
|
# We do not want to mix coordinator lock with the user lock not to create cases where
|
@@ -36,11 +37,11 @@ module Karafka
|
|
36
37
|
# framework and can be used for user-facing locking
|
37
38
|
@shared_mutex = Mutex.new
|
38
39
|
@collapser = Collapser.new
|
39
|
-
@filter = FiltersApplier.new(self)
|
40
|
+
@filter = Coordinators::FiltersApplier.new(self)
|
40
41
|
|
41
42
|
return unless topic.virtual_partitions?
|
42
43
|
|
43
|
-
@virtual_offset_manager = VirtualOffsetManager.new(
|
44
|
+
@virtual_offset_manager = Coordinators::VirtualOffsetManager.new(
|
44
45
|
topic.name,
|
45
46
|
partition,
|
46
47
|
topic.virtual_partitions.offset_metadata_strategy
|
@@ -64,6 +65,14 @@ module Karafka
|
|
64
65
|
|
65
66
|
@filter.apply!(messages)
|
66
67
|
|
68
|
+
# Do not clear coordinator errors storage when we are retrying, so we can reference the
|
69
|
+
# errors that have happened during recovery. This can be useful for implementing custom
|
70
|
+
# flows. There can be more errors than one when running with virtual partitions so we
|
71
|
+
# need to make sure we collect them all. Under collapse when we reference a given
|
72
|
+
# consumer we should be able to get all the errors and not just first/last.
|
73
|
+
#
|
74
|
+
# @note We use zero as the attempt mark because we are not "yet" in the attempt 1
|
75
|
+
@errors_tracker.clear if attempt.zero?
|
67
76
|
@executed.clear
|
68
77
|
|
69
78
|
# We keep the old processed offsets until the collapsing is done and regular processing
|
@@ -79,6 +88,7 @@ module Karafka
|
|
79
88
|
# @param error [StandardError] error from the failure
|
80
89
|
def failure!(consumer, error)
|
81
90
|
super
|
91
|
+
@errors_tracker << error
|
82
92
|
collapse_until!(@last_message.offset + 1)
|
83
93
|
end
|
84
94
|
|
@@ -0,0 +1,74 @@
|
|
1
|
+
# frozen_string_literal: true
|
2
|
+
|
3
|
+
# This Karafka component is a Pro component under a commercial license.
|
4
|
+
# This Karafka component is NOT licensed under LGPL.
|
5
|
+
#
|
6
|
+
# All of the commercial components are present in the lib/karafka/pro directory of this
|
7
|
+
# repository and their usage requires commercial license agreement.
|
8
|
+
#
|
9
|
+
# Karafka has also commercial-friendly license, commercial support and commercial components.
|
10
|
+
#
|
11
|
+
# By sending a pull request to the pro components, you are agreeing to transfer the copyright of
|
12
|
+
# your code to Maciej Mensfeld.
|
13
|
+
|
14
|
+
module Karafka
|
15
|
+
module Pro
|
16
|
+
module Processing
|
17
|
+
# Namespace for Pro coordinator related sub-components
|
18
|
+
module Coordinators
|
19
|
+
# Object used to track errors in between executions to be able to build error-type based
|
20
|
+
# recovery flows.
|
21
|
+
class ErrorsTracker
|
22
|
+
include Enumerable
|
23
|
+
|
24
|
+
# Max errors we keep in memory.
|
25
|
+
# We do not want to keep more because for DLQ-less this would cause memory-leaks.
|
26
|
+
STORAGE_LIMIT = 100
|
27
|
+
|
28
|
+
private_constant :STORAGE_LIMIT
|
29
|
+
|
30
|
+
def initialize
|
31
|
+
@errors = []
|
32
|
+
end
|
33
|
+
|
34
|
+
# Clears all the errors
|
35
|
+
def clear
|
36
|
+
@errors.clear
|
37
|
+
end
|
38
|
+
|
39
|
+
# @param error [StandardError] adds the error to the tracker
|
40
|
+
def <<(error)
|
41
|
+
@errors.shift if @errors.size >= STORAGE_LIMIT
|
42
|
+
@errors << error
|
43
|
+
end
|
44
|
+
|
45
|
+
# @return [Boolean] is the error tracker empty
|
46
|
+
def empty?
|
47
|
+
@errors.empty?
|
48
|
+
end
|
49
|
+
|
50
|
+
# @return [Integer] number of elements
|
51
|
+
def size
|
52
|
+
count
|
53
|
+
end
|
54
|
+
|
55
|
+
# @return [StandardError, nil] last error that occurred or nil if no errors
|
56
|
+
def last
|
57
|
+
@errors.last
|
58
|
+
end
|
59
|
+
|
60
|
+
# Iterates over errors
|
61
|
+
# @param block [Proc] code we want to run on each error
|
62
|
+
def each(&block)
|
63
|
+
@errors.each(&block)
|
64
|
+
end
|
65
|
+
|
66
|
+
# @return [Array<StandardError>] array with all the errors that occurred
|
67
|
+
def all
|
68
|
+
@errors
|
69
|
+
end
|
70
|
+
end
|
71
|
+
end
|
72
|
+
end
|
73
|
+
end
|
74
|
+
end
|
@@ -0,0 +1,107 @@
|
|
1
|
+
# frozen_string_literal: true
|
2
|
+
|
3
|
+
# This Karafka component is a Pro component under a commercial license.
|
4
|
+
# This Karafka component is NOT licensed under LGPL.
|
5
|
+
#
|
6
|
+
# All of the commercial components are present in the lib/karafka/pro directory of this
|
7
|
+
# repository and their usage requires commercial license agreement.
|
8
|
+
#
|
9
|
+
# Karafka has also commercial-friendly license, commercial support and commercial components.
|
10
|
+
#
|
11
|
+
# By sending a pull request to the pro components, you are agreeing to transfer the copyright of
|
12
|
+
# your code to Maciej Mensfeld.
|
13
|
+
|
14
|
+
module Karafka
|
15
|
+
module Pro
|
16
|
+
module Processing
|
17
|
+
module Coordinators
|
18
|
+
# Applier for all filters we want to have. Whether related to limiting messages based
|
19
|
+
# on the payload or any other things.
|
20
|
+
#
|
21
|
+
# From the outside world perspective, this encapsulates all the filters.
|
22
|
+
# This means that this is the API we expose as a single filter, allowing us to control
|
23
|
+
# the filtering via many filters easily.
|
24
|
+
class FiltersApplier
|
25
|
+
# @return [Array] registered filters array. Useful if we want to inject internal context
|
26
|
+
# aware filters.
|
27
|
+
attr_reader :filters
|
28
|
+
|
29
|
+
# @param coordinator [Pro::Coordinator] pro coordinator
|
30
|
+
def initialize(coordinator)
|
31
|
+
# Builds filters out of their factories
|
32
|
+
# We build it that way (providing topic and partition) because there may be a case
|
33
|
+
# where someone wants to have a specific logic that is per topic or partition. Like for
|
34
|
+
# example a case where there is a cache bypassing revocations for topic partition.
|
35
|
+
#
|
36
|
+
# We provide full Karafka routing topic here and not the name only, in case the filter
|
37
|
+
# would be customized based on other topic settings (like VPs, etc)
|
38
|
+
#
|
39
|
+
# This setup allows for biggest flexibility also because topic object holds the
|
40
|
+
# reference to the subscription group and consumer group
|
41
|
+
@filters = coordinator.topic.filtering.factories.map do |factory|
|
42
|
+
factory.call(coordinator.topic, coordinator.partition)
|
43
|
+
end
|
44
|
+
end
|
45
|
+
|
46
|
+
# @param messages [Array<Karafka::Messages::Message>] array with messages from the
|
47
|
+
# partition
|
48
|
+
def apply!(messages)
|
49
|
+
return unless active?
|
50
|
+
|
51
|
+
@filters.each { |filter| filter.apply!(messages) }
|
52
|
+
end
|
53
|
+
|
54
|
+
# @return [Boolean] did we filter out any messages during filtering run
|
55
|
+
def applied?
|
56
|
+
return false unless active?
|
57
|
+
|
58
|
+
!applied.empty?
|
59
|
+
end
|
60
|
+
|
61
|
+
# @return [Symbol] consumer post-filtering action that should be taken
|
62
|
+
def action
|
63
|
+
return :skip unless applied?
|
64
|
+
|
65
|
+
# The highest priority is on a potential backoff from any of the filters because it is
|
66
|
+
# the less risky (delay and continue later)
|
67
|
+
return :pause if applied.any? { |filter| filter.action == :pause }
|
68
|
+
|
69
|
+
# If none of the filters wanted to pause, we can check for any that would want to seek
|
70
|
+
# and if there is any, we can go with this strategy
|
71
|
+
return :seek if applied.any? { |filter| filter.action == :seek }
|
72
|
+
|
73
|
+
:skip
|
74
|
+
end
|
75
|
+
|
76
|
+
# @return [Integer] minimum timeout we need to pause. This is the minimum for all the
|
77
|
+
# filters to satisfy all of them.
|
78
|
+
def timeout
|
79
|
+
applied.map(&:timeout).compact.min || 0
|
80
|
+
end
|
81
|
+
|
82
|
+
# The first message we do need to get next time we poll. We use the minimum not to jump
|
83
|
+
# accidentally by over any.
|
84
|
+
# @return [Karafka::Messages::Message, nil] cursor message or nil if none
|
85
|
+
# @note Cursor message can also return the offset in the time format
|
86
|
+
def cursor
|
87
|
+
return nil unless active?
|
88
|
+
|
89
|
+
applied.map(&:cursor).compact.min_by(&:offset)
|
90
|
+
end
|
91
|
+
|
92
|
+
private
|
93
|
+
|
94
|
+
# @return [Boolean] is filtering active
|
95
|
+
def active?
|
96
|
+
!@filters.empty?
|
97
|
+
end
|
98
|
+
|
99
|
+
# @return [Array<Object>] filters that applied any sort of messages limiting
|
100
|
+
def applied
|
101
|
+
@filters.select(&:applied?)
|
102
|
+
end
|
103
|
+
end
|
104
|
+
end
|
105
|
+
end
|
106
|
+
end
|
107
|
+
end
|