vitess.io/vitess@v0.16.2/go/vt/vtgate/buffer/variables.go (about)

     1  /*
     2  Copyright 2019 The Vitess Authors.
     3  
     4  Licensed under the Apache License, Version 2.0 (the "License");
     5  you may not use this file except in compliance with the License.
     6  You may obtain a copy of the License at
     7  
     8      http://www.apache.org/licenses/LICENSE-2.0
     9  
    10  Unless required by applicable law or agreed to in writing, software
    11  distributed under the License is distributed on an "AS IS" BASIS,
    12  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    13  See the License for the specific language governing permissions and
    14  limitations under the License.
    15  */
    16  
    17  package buffer
    18  
    19  import (
    20  	"vitess.io/vitess/go/stats"
    21  )
    22  
    23  // This file contains all status variables which can be used to monitor the
    24  // buffer.
    25  
    26  var (
    27  	// starts counts how often we started buffering (including dry-run bufferings).
    28  	starts = stats.NewCountersWithMultiLabels(
    29  		"BufferStarts",
    30  		"Buffering operation starts, including dry-run",
    31  		[]string{"Keyspace", "ShardName"})
    32  	// stops counts how often we triggered the stop of a buffering, including
    33  	// dry-run bufferings.
    34  	// See the type "stopReason" below for all possible values of "Reason".
    35  	stops = stats.NewCountersWithMultiLabels(
    36  		"BufferStops",
    37  		"Buffering operation stops, including dry-runs",
    38  		[]string{"Keyspace", "ShardName", "Reason"})
    39  
    40  	// failoverDurationSumMs is the cumulative sum of all failover durations.
    41  	// In connection with "starts" it can be used to calculate a moving average.
    42  	failoverDurationSumMs = stats.NewCountersWithMultiLabels(
    43  		"BufferFailoverDurationSumMs",
    44  		"Total buffering failover duration",
    45  		[]string{"Keyspace", "ShardName"})
    46  
    47  	// utilizationSum is the cumulative sum of the maximum buffer utilization
    48  	// (in percentage) during each failover.
    49  	// Utilization = maximum number of requests buffered / buffer size.
    50  	// In connection with "starts" it can be used to calculate a moving average.
    51  	// TODO(mberlin): Replace this with a MultiHistogram once it's available.
    52  	utilizationSum = stats.NewGaugesWithMultiLabels(
    53  		"BufferUtilizationSum",
    54  		"Cumulative buffer utilization (in %) during failover",
    55  		[]string{"Keyspace", "ShardName"})
    56  	// utilizationDryRunSum is the cumulative sum of the maximum *theoretical*
    57  	// buffer utilization (in percentage) during each failover.
    58  	// Utilization = maximum number of requests buffered seen / buffer size.
    59  	// In connection with "starts" it can be used to calculate a moving average.
    60  	// Example: Buffer size = 10. Two consecutive failovers with maximum values of
    61  	// 15 and 5 seen requests each add up to a value of 200% (150% + 50%
    62  	// utilization). The moving average would be 100% because there were two
    63  	// failovers in that period.
    64  	// TODO(mberlin): Replace this with a MultiHistogram once it's available.
    65  	utilizationDryRunSum = stats.NewCountersWithMultiLabels(
    66  		"BufferUtilizationDryRunSum",
    67  		"Cumulative buffer utilization % during failover (dry-run)",
    68  		[]string{"Keyspace", "ShardName"})
    69  
    70  	// requestsBuffered tracks how many requests were added to the buffer.
    71  	// NOTE: The two counters "Buffered" and "Skipped" should cover all requests
    72  	// which passed through the buffer.
    73  	requestsBuffered = stats.NewCountersWithMultiLabels(
    74  		"BufferRequestsBuffered",
    75  		"Buffered requests",
    76  		[]string{"Keyspace", "ShardName"})
    77  	// requestsBufferedDryRun tracks how many requests would have been added to
    78  	// the buffer (dry-run mode).
    79  	requestsBufferedDryRun = stats.NewCountersWithMultiLabels(
    80  		"BufferRequestsBufferedDryRun",
    81  		"Buffered requests (dry-run)",
    82  		[]string{"Keyspace", "ShardName"})
    83  	// requestsBuffered tracks how many requests were drained from the buffer.
    84  	// NOTE: The sum of the two counters "Drained" and "Evicted" should be
    85  	// identical to the "Buffered" counter value.
    86  	requestsDrained = stats.NewCountersWithMultiLabels(
    87  		"BufferRequestsDrained",
    88  		"Drained buffered requests",
    89  		[]string{"Keyspace", "ShardName"})
    90  	// requestsEvicted tracks how many requests were evicted early from the buffer.
    91  	// See the type "evictedReason" below for all possible values of "Reason".
    92  	requestsEvicted = stats.NewCountersWithMultiLabels(
    93  		"BufferRequestsEvicted",
    94  		"Evicted buffered requests",
    95  		[]string{"Keyspace", "ShardName", "Reason"})
    96  	// requestsSkipped tracks how many requests would have been buffered but
    97  	// eventually were not (includes dry-run bufferings).
    98  	// See the type "skippedReason" below for all possible values of "Reason".
    99  	requestsSkipped = stats.NewCountersWithMultiLabels(
   100  		"BufferRequestsSkipped",
   101  		"Skipped buffering requests (incl. dry-run)",
   102  		[]string{"Keyspace", "ShardName", "Reason"})
   103  )
   104  
   105  // stopReason is used in "stopsByReason" as "Reason" label.
   106  type stopReason string
   107  
   108  var stopReasons = []stopReason{stopShardMissing, stopFailoverEndDetected, stopMaxFailoverDurationExceeded, stopShutdown}
   109  
   110  const (
   111  	stopShardMissing                stopReason = "ReshardingComplete"
   112  	stopFailoverEndDetected         stopReason = "NewPrimarySeen"
   113  	stopMaxFailoverDurationExceeded stopReason = "MaxDurationExceeded"
   114  	stopShutdown                    stopReason = "Shutdown"
   115  )
   116  
   117  // evictedReason is used in "requestsEvicted" as "Reason" label.
   118  type evictedReason string
   119  
   120  var evictReasons = []evictedReason{evictedContextDone, evictedBufferFull, evictedWindowExceeded}
   121  
   122  const (
   123  	evictedContextDone evictedReason = "ContextDone"
   124  	//lint: ignore SA9004 ok not to use explicit type here because implicit type string is correct
   125  	evictedBufferFull     = "BufferFull"
   126  	evictedWindowExceeded = "WindowExceeded"
   127  )
   128  
   129  // skippedReason is used in "requestsSkipped" as "Reason" label.
   130  type skippedReason string
   131  
   132  var skippedReasons = []skippedReason{skippedBufferFull, skippedDisabled, skippedShutdown, skippedLastReparentTooRecent, skippedLastFailoverTooRecent}
   133  
   134  const (
   135  	// skippedBufferFull occurs when all slots in the buffer are occupied by one
   136  	// or more concurrent failovers. Unlike "evictedBufferFull", no request could
   137  	// be evicted and therefore we had to skip this request.
   138  	skippedBufferFull skippedReason = "BufferFull"
   139  	// skippedDisabled is used when the buffer was disabled for that particular
   140  	// keyspace/shard.
   141  	skippedDisabled              = "Disabled"
   142  	skippedShutdown              = "Shutdown"
   143  	skippedLastReparentTooRecent = "LastReparentTooRecent"
   144  	skippedLastFailoverTooRecent = "LastFailoverTooRecent"
   145  )
   146  
   147  // initVariablesForShard is used to initialize all shard variables to 0.
   148  // If we don't do this, monitoring frameworks may not correctly calculate rates
   149  // for the first failover of the shard because they see a transition from
   150  // "no value for this label set (NaN)" to "a value".
   151  // "statsKey" should have two members for keyspace and shard.
   152  func initVariablesForShard(statsKey []string) {
   153  	starts.Reset(statsKey)
   154  	for _, reason := range stopReasons {
   155  		key := append(statsKey, string(reason))
   156  		stops.Reset(key)
   157  	}
   158  
   159  	failoverDurationSumMs.Reset(statsKey)
   160  
   161  	utilizationSum.Set(statsKey, 0)
   162  	utilizationDryRunSum.Reset(statsKey)
   163  
   164  	requestsBuffered.Reset(statsKey)
   165  	requestsBufferedDryRun.Reset(statsKey)
   166  	requestsDrained.Reset(statsKey)
   167  	for _, reason := range evictReasons {
   168  		key := append(statsKey, string(reason))
   169  		requestsEvicted.Reset(key)
   170  	}
   171  	for _, reason := range skippedReasons {
   172  		key := append(statsKey, string(reason))
   173  		requestsSkipped.Reset(key)
   174  	}
   175  }
   176  
   177  // TODO(mberlin): Remove the gauge values below once we store them
   178  // internally and have a /bufferz page where we can show this.
   179  var (
   180  	// bufferSizeStat publishes the configured per vtgate buffer size. It can be used
   181  	// to calculate the utilization of the buffer.
   182  	bufferSizeStat         = stats.NewGauge("BufferSize", "The configured per vtgate buffer size")
   183  	lastFailoverDurationMs = stats.NewGaugesWithMultiLabels(
   184  		"BufferLastFailoverDurationMs",
   185  		"Buffered requests during the last failover. The value for a given shard will be reset at the next failover.",
   186  		[]string{"Keyspace", "ShardName"})
   187  	lastRequestsInFlightMax = stats.NewGaugesWithMultiLabels(
   188  		"BufferLastRequestsInFlightMax",
   189  		"The max value of buffered requests in flight of the last failover. The value for a given shard will be reset at the next failover.",
   190  		[]string{"Keyspace", "ShardName"})
   191  	// lastRequestsDryRunMax has the maximum number of requests which were seen during
   192  	// a dry-run buffering of the last failover.
   193  	// The value for a given shard will be reset at the next failover.
   194  	lastRequestsDryRunMax = stats.NewGaugesWithMultiLabels(
   195  		"BufferLastRequestsDryRunMax",
   196  		"Max # of requests which were seen during a dry-run buffering of the last failover",
   197  		[]string{"Keyspace", "ShardName"})
   198  )