Module: KubernetesMetadata::WatchPods

Includes:
Common
Included in:
Fluent::Plugin::KubernetesMetadataFilter
Defined in:
lib/fluent/plugin/kubernetes_metadata_watch_pods.rb

Instance Method Summary collapse

Methods included from Common

#match_annotations, #parse_namespace_metadata, #parse_pod_metadata, #syms_to_strs

Instance Method Details

#get_pods_and_start_watcher ⇒ Object

List all pods, record the resourceVersion and return a watcher starting from that resourceVersion.



125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
 
# File 'lib/fluent/plugin/kubernetes_metadata_watch_pods.rb', line 125

def get_pods_and_start_watcher
  options = {
    resource_version: '0' # Fetch from API server cache instead of etcd quorum read
  }
  if ENV['K8S_NODE_NAME']
    options[:field_selector] = 'spec.nodeName=' + ENV['K8S_NODE_NAME']
  end
  if @last_seen_resource_version
    options[:resource_version] = @last_seen_resource_version
  else
    pods = @client.get_pods(options)
    pods[:items].each do |pod|
      cache_key = pod[:metadata][:uid]
      @cache[cache_key] = parse_pod_metadata(pod)
      @stats.bump(:pod_cache_host_updates)
    end

    # continue watching from most recent resourceVersion
    options[:resource_version] = pods[:metadata][:resourceVersion]
  end

  watcher = @client.watch_pods(options)
  reset_pod_watch_retry_stats
  watcher
end

#process_pod_watcher_notices(watcher) ⇒ Object

Process a watcher notice and potentially raise an exception.



159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
 
# File 'lib/fluent/plugin/kubernetes_metadata_watch_pods.rb', line 159

def process_pod_watcher_notices(watcher)
  watcher.each do |notice|
    # store version we processed to not reprocess it ... do not unset when there is no version in response
    version = ( # TODO: replace with &.dig once we are on ruby 2.5+
      notice[:object] && notice[:object][:metadata] && notice[:object][:metadata][:resourceVersion]
    )
    @last_seen_resource_version = version if version

    case notice[:type]
    when 'MODIFIED'
      reset_pod_watch_retry_stats
      cache_key = notice.dig(:object, :metadata, :uid)
      cached    = @cache[cache_key]
      if cached
        @cache[cache_key] = parse_pod_metadata(notice[:object])
        @stats.bump(:pod_cache_watch_updates)
      elsif ENV['K8S_NODE_NAME'] == notice[:object][:spec][:nodeName]
        @cache[cache_key] = parse_pod_metadata(notice[:object])
        @stats.bump(:pod_cache_host_updates)
      else
        @stats.bump(:pod_cache_watch_misses)
      end
    when 'DELETED'
      reset_pod_watch_retry_stats
      # ignore and let age out for cases where pods
      # deleted but still processing logs
      @stats.bump(:pod_cache_watch_delete_ignored)
    when 'ERROR'
      if notice[:object] && notice[:object][:code] == 410
        @last_seen_resource_version = nil # requested resourceVersion was too old, need to reset
        @stats.bump(:pod_watch_gone_notices)
        raise GoneError
      else
        @stats.bump(:pod_watch_error_type_notices)
        message = notice[:object][:message] if notice[:object] && notice[:object][:message]
        raise "Error while watching pods: #{message}"
      end
    else
      reset_pod_watch_retry_stats
      # Don't pay attention to creations, since the created pod may not
      # end up on this node.
      @stats.bump(:pod_cache_watch_ignored)
    end
  end
end

#reset_pod_watch_retry_stats ⇒ Object

Reset pod watch retry count and backoff interval as there is a successful watch notice.



153
154
155
156
 
# File 'lib/fluent/plugin/kubernetes_metadata_watch_pods.rb', line 153

def reset_pod_watch_retry_stats
  Thread.current[:pod_watch_retry_count] = 0
  Thread.current[:pod_watch_retry_backoff_interval] = @watch_retry_interval
end

#set_up_pod_thread ⇒ Object 



28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
 
# File 'lib/fluent/plugin/kubernetes_metadata_watch_pods.rb', line 28

def set_up_pod_thread
  # Any failures / exceptions in the initial setup should raise
  # Fluent:ConfigError, so that users can inspect potential errors in
  # the configuration.
  pod_watcher = start_pod_watch

  Thread.current[:pod_watch_retry_backoff_interval] = @watch_retry_interval
  Thread.current[:pod_watch_retry_count] = 0

  # Any failures / exceptions in the followup watcher notice
  # processing will be swallowed and retried. These failures /
  # exceptions could be caused by Kubernetes API being temporarily
  # down. We assume the configuration is correct at this point.
  loop do
    pod_watcher ||= get_pods_and_start_watcher
    process_pod_watcher_notices(pod_watcher)
  rescue GoneError => e
    # Expected error. Quietly go back through the loop in order to
    # start watching from the latest resource versions
    @stats.bump(:pod_watch_gone_errors)
    log.info('410 Gone encountered. Restarting pod watch to reset resource versions.', e)
    pod_watcher = nil
  rescue KubeException => e
    if e.error_code == 401
      # recreate client to refresh token
      log.info("Encountered '401 Unauthorized' exception in watch, recreating client to refresh token")
      create_client()
      pod_watcher = nil
    else
      # treat all other errors the same as StandardError, log, swallow and reset
      @stats.bump(:pod_watch_failures)
      if Thread.current[:pod_watch_retry_count] < @watch_retry_max_times
        # Instead of raising exceptions and crashing Fluentd, swallow
        # the exception and reset the watcher.
        log.info(
          'Exception encountered parsing pod watch event. The ' \
          'connection might have been closed. Sleeping for ' \
          "#{Thread.current[:pod_watch_retry_backoff_interval]} " \
          'seconds and resetting the pod watcher.', e
        )
        sleep(Thread.current[:pod_watch_retry_backoff_interval])
        Thread.current[:pod_watch_retry_count] += 1
        Thread.current[:pod_watch_retry_backoff_interval] *= @watch_retry_exponential_backoff_base
        pod_watcher = nil
      else
        # Since retries failed for many times, log as errors instead
        # of info and raise exceptions and trigger Fluentd to restart.
        message =
          'Exception encountered parsing pod watch event. The ' \
          'connection might have been closed. Retried ' \
          "#{@watch_retry_max_times} times yet still failing. Restarting."
        log.error(message, e)
        raise Fluent::UnrecoverableError, message
      end
    end
  rescue StandardError => e
    @stats.bump(:pod_watch_failures)
    if Thread.current[:pod_watch_retry_count] < @watch_retry_max_times
      # Instead of raising exceptions and crashing Fluentd, swallow
      # the exception and reset the watcher.
      log.info(
        'Exception encountered parsing pod watch event. The ' \
        'connection might have been closed. Sleeping for ' \
        "#{Thread.current[:pod_watch_retry_backoff_interval]} " \
        'seconds and resetting the pod watcher.', e
      )
      sleep(Thread.current[:pod_watch_retry_backoff_interval])
      Thread.current[:pod_watch_retry_count] += 1
      Thread.current[:pod_watch_retry_backoff_interval] *= @watch_retry_exponential_backoff_base
      pod_watcher = nil
    else
      # Since retries failed for many times, log as errors instead
      # of info and raise exceptions and trigger Fluentd to restart.
      message =
        'Exception encountered parsing pod watch event. The ' \
        'connection might have been closed. Retried ' \
        "#{@watch_retry_max_times} times yet still failing. Restarting."
      log.error(message, e)
      raise Fluent::UnrecoverableError, message
    end
  end
end

#start_pod_watch ⇒ Object 



111
112
113
114
115
116
117
118
119
120
121
 
# File 'lib/fluent/plugin/kubernetes_metadata_watch_pods.rb', line 111

def start_pod_watch
  get_pods_and_start_watcher
rescue StandardError => e
  message = 'start_pod_watch: Exception encountered setting up pod watch ' \
            "from Kubernetes API #{@apiVersion} endpoint " \
            "#{@kubernetes_url}: #{e.message}"
  message += " (#{e.response})" if e.respond_to?(:response)
  log.debug(message)

  raise Fluent::ConfigError, message
end