network_resiliency 0.7.9 → 0.7.11

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: a284503d2243c5b00f5ab5fc09ca842667ebb6db7ab489fd8fa0ce033b2e4b66
-  data.tar.gz: bcf55039bcfa3e4969a7d1a26f3393b362dd64b7318d0322b0cbe1755ce7c0f7
+  metadata.gz: 1eec5786e7daaea4b6f3c285dffc1e4bf1c70d7096ce69d44f4cc4ea80111ea5
+  data.tar.gz: bc3ca5e592f1a809bfb28453e43d49a64405a424861aef2c6c6745ef2f27ebb4
 SHA512:
-  metadata.gz: 6a40fb761f8b03e3af86afcf9c37247b6e8666bd5326fb5d8922ed35b78d9fd95c700de8a4ce142d014143d55eaf4773dbc152a56dea031400151cd265dcaba7
-  data.tar.gz: f74ef650a917987afbcb72133f6b26491963cac0cb595f5f2505f49c70b0aa67dd5c1db8f290f3ad6a3708a4d950397111b5c487934082d403e2cd7f41e99cd1
+  metadata.gz: fa1af69b1f175fb708247ddd0e9501ae2b945909197b26c343347e113bb78e3e645c5217c5f74d63fda780a027b730746b8641b9f77cb5c31626bc2f985c60ff
+  data.tar.gz: e9733ddd7978b321f8c49218de503f42b300aa269757559cac442dcefb43e0b136e1507360629e9aad1976cedd31c7c0ccd4e9a256c41962a990e50f046d3979

data/CHANGELOG.md

@@ -1,3 +1,14 @@
+###  v0.7.11  (2024-02-20)
+- lower dynamic timeout
+- fix metric specs
+- extend cache ttl
+
+###  v0.7.10  (2024-02-15)
+- ddog sampling
+- reconnect redis
+- sync logging
+- warn method
+
 ###  v0.7.9  (2024-02-07)
 - sync frequency
 

data/Gemfile.lock

@@ -1,7 +1,7 @@
 PATH
   remote: .
   specs:
-    network_resiliency (0.7.9)
+    network_resiliency (0.7.11)
 
 GEM
   remote: https://rubygems.org/

data/lib/network_resiliency/stats.rb

@@ -97,7 +97,7 @@ module NetworkResiliency
     MIN_SAMPLE_SIZE = 1000
     MAX_WINDOW_LENGTH = 1000
     STATS_TTL = 24 * 60 * 60 # 1 day
-    CACHE_TTL = 60 # seconds
+    CACHE_TTL = 120 # seconds
 
     LUA_SCRIPT = <<~LUA
       local results = {}
@@ -189,11 +189,9 @@ module NetworkResiliency
       end
 
       res = redis.eval(LUA_SCRIPT, keys, args)
-      data.keys.zip(res.each_slice(3)).map do |key, stats|
-        n, avg, sq_dist = *stats
-
-        [ key, Stats.from(n: n, avg: avg, sq_dist: sq_dist) ]
-      end.to_h
+      data.keys.zip(res.each_slice(3)).to_h.transform_values! do |n, avg, sq_dist|
+        Stats.from(n: n, avg: avg, sq_dist: sq_dist)
+      end
     end
 
     def self.fetch(redis, keys)

data/lib/network_resiliency/stats_engine.rb

@@ -65,11 +65,13 @@ module NetworkResiliency
           empty: data.empty?,
           truncated: data.size < dirty_keys.size,
         }.select { |_, v| v },
+        sample_rate: SAMPLE_RATE[:sync],
       )
 
       NetworkResiliency.statsd&.distribution(
         "network_resiliency.sync.keys.dirty",
         dirty_keys.select { |_, n| n > 0 }.count,
+        sample_rate: SAMPLE_RATE[:sync],
       )
 
       return [] if data.empty?

data/lib/network_resiliency/syncer.rb

@@ -1,6 +1,7 @@
 module NetworkResiliency
   class Syncer < Thread
     LOCK = Mutex.new
+    SLEEP_DURATION = 10
 
     class << self
       def start
@@ -45,12 +46,17 @@ module NetworkResiliency
     private
 
     def sync
+      # force redis to reconnect post fork
+      NetworkResiliency.redis.disconnect! if NetworkResiliency.redis.connected?
+
       until @shutdown
         StatsEngine.sync(NetworkResiliency.redis)
 
-        sleep(10)
+        sleep(SLEEP_DURATION)
       end
     rescue Interrupt
+    rescue => e
+      NetworkResiliency.warn(__method__, e)
     end
   end
 end

data/lib/network_resiliency/version.rb

@@ -1,3 +1,3 @@
 module NetworkResiliency
-  VERSION = "0.7.9"
+  VERSION = "0.7.11"
 end

data/lib/network_resiliency.rb

@@ -20,6 +20,11 @@ module NetworkResiliency
   ADAPTERS = [ :http, :faraday, :redis, :mysql, :postgres, :rails ].freeze
   MODE = [ :observe, :resilient ].freeze
   RESILIENCY_SIZE_THRESHOLD = 1_000
+  SAMPLE_RATE = {
+    timeout: 0.1,
+    stats: 0.1,
+    sync: 0.1,
+  }
 
   extend self
 
@@ -121,15 +126,7 @@ module NetworkResiliency
 
     mode
   rescue => e
-    NetworkResiliency.statsd&.increment(
-      "network_resiliency.error",
-      tags: {
-        method: __method__,
-        type: e.class,
-      },
-    )
-
-    warn "[ERROR] NetworkResiliency: #{e.class}: #{e.message}"
+    warn(__method__, e)
 
     :observe
   end
@@ -243,6 +240,7 @@ module NetworkResiliency
         adapter: adapter,
         destination: destination,
       },
+      sample_rate: SAMPLE_RATE[:timeout],
     ) if timeout && timeout > 0
 
     if error
@@ -272,32 +270,27 @@ module NetworkResiliency
         "network_resiliency.#{action}.stats.n",
         stats.n,
         tags: tags,
+        sample_rate: SAMPLE_RATE[:stats],
       )
 
       NetworkResiliency.statsd&.distribution(
         "network_resiliency.#{action}.stats.avg",
         stats.avg,
         tags: tags,
+        sample_rate: SAMPLE_RATE[:stats],
       )
 
       NetworkResiliency.statsd&.distribution(
         "network_resiliency.#{action}.stats.stdev",
         stats.stdev,
         tags: tags,
+        sample_rate: SAMPLE_RATE[:stats],
       )
     end
 
     nil
   rescue => e
-    NetworkResiliency.statsd&.increment(
-      "network_resiliency.error",
-      tags: {
-        method: __method__,
-        type: e.class,
-      },
-    )
-
-    warn "[ERROR] NetworkResiliency: #{e.class}: #{e.message}"
+    warn(__method__, e)
   end
 
   IP_ADDRESS_REGEX = /\d{1,3}(\.\d{1,3}){3}/
@@ -323,7 +316,7 @@ module NetworkResiliency
       destination: destination,
     }
 
-    p99 = (stats.avg + stats.stdev * 3).order_of_magnitude(ceil: true)
+    p99 = (stats.avg + stats.stdev * 2).order_of_magnitude(ceil: true)
 
     timeouts = []
 
@@ -343,6 +336,7 @@ module NetworkResiliency
           NetworkResiliency.statsd&.increment(
             "network_resiliency.timeout.raised",
             tags: tags,
+            sample_rate: SAMPLE_RATE[:timeout],
           )
         end
       else
@@ -352,6 +346,7 @@ module NetworkResiliency
         NetworkResiliency.statsd&.increment(
           "network_resiliency.timeout.too_low",
           tags: tags,
+          sample_rate: SAMPLE_RATE[:timeout],
         )
       end
     else
@@ -365,6 +360,7 @@ module NetworkResiliency
       NetworkResiliency.statsd&.increment(
         "network_resiliency.timeout.missing",
         tags: tags,
+        sample_rate: SAMPLE_RATE[:timeout],
       )
     end
 
@@ -375,6 +371,7 @@ module NetworkResiliency
         adapter: adapter,
         destination: destination,
       },
+      sample_rate: SAMPLE_RATE[:timeout],
     )
 
     case units
@@ -386,15 +383,7 @@ module NetworkResiliency
       raise ArgumentError, "invalid units: #{units}"
     end
   rescue => e
-    NetworkResiliency.statsd&.increment(
-      "network_resiliency.error",
-      tags: {
-        method: __method__,
-        type: e.class,
-      },
-    )
-
-    warn "[ERROR] NetworkResiliency: #{e.class}: #{e.message}"
+    warn(__method__, e)
 
     default
   end
@@ -414,4 +403,16 @@ module NetworkResiliency
   def thread_state
     Thread.current["network_resiliency"] ||= {}
   end
+
+  def warn(method, e)
+    NetworkResiliency.statsd&.increment(
+      "network_resiliency.error",
+      tags: {
+        method: method,
+        type: e.class,
+      },
+    )
+
+    Kernel.warn "[ERROR] NetworkResiliency #{method}: #{e.class}: #{e.message}"
+  end
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: network_resiliency
 version: !ruby/object:Gem::Version
-  version: 0.7.9
+  version: 0.7.11
 platform: ruby
 authors:
 - Daniel Pepper
 autorequire:
 bindir: bin
 cert_chain: []
-date: 2024-02-08 00:00:00.000000000 Z
+date: 2024-02-20 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: byebug