prometheus/retrieval/scheduler.go
Matt T. Proud efe61c18fa Refactor target scheduling to separate facility.
``Target`` will be refactored down the road to support various
nuanced endpoint types.  Thusly incorporating the scheduling
behavior within it will be problematic.  To that end, the scheduling
behavior has been moved into a separate assistance type to improve
conciseness and testability.

``make format`` was also run.
2013-01-13 10:43:37 +01:00

144 lines
5 KiB
Go

// Copyright 2013 Prometheus Team
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package retrieval
import (
"math"
"time"
)
const (
// The default increment for exponential backoff when querying a target.
DEFAULT_BACKOFF_VALUE = 2
// The base units for the exponential backoff.
DEFAULT_BACKOFF_VALUE_UNIT = time.Second
// The maximum allowed backoff time.
MAXIMUM_BACKOFF_VALUE = 30 * time.Minute
)
// A basic interface only useful in testing contexts for dispensing the time
// in a controlled manner.
type instantProvider interface {
// The current instant.
Now() time.Time
}
// timer is a simple means for fluently wrapping around standard Go timekeeping
// mechanisms to enhance testability without compromising code readability.
//
// A timer is sufficient for use on bare initialization. A provider should be
// set only for test contexts. When not provided, a timer emits the current
// system time.
type timer struct {
// The underlying means through which time is provided, if supplied.
provider instantProvider
}
// Emit the current instant.
func (t timer) Now() time.Time {
if t.provider == nil {
return time.Now()
}
return t.provider.Now()
}
// scheduler is an interface that various scheduling strategies must fulfill
// in order to set the scheduling order for a target.
//
// Target takes advantage of this type by embedding an instance of scheduler
// in each Target instance itself. The emitted scheduler.ScheduledFor() is
// the basis for sorting the order of pending queries.
//
// This type is described as an interface to maximize testability.
type scheduler interface {
// ScheduledFor emits the earliest time at which the given object is allowed
// to be run. This time may or not be a reflection of the earliest parameter
// provided in Reschedule; that is up to the underlying strategy
// implementations.
ScheduledFor() time.Time
// Instruct the scheduled item to re-schedule itself given new state data and
// the earliest time at which the outside system thinks the operation should
// be scheduled for.
Reschedule(earliest time.Time, future TargetState)
}
// healthScheduler is an implementation of scheduler that uses health data
// provided by the target field as well as unreachability counts to determine
// when to next schedule an operation.
//
// The type is almost capable of being used with default initialization, except
// that a target field must be provided for which the system compares current
// health against future proposed values.
type healthScheduler struct {
scheduledFor time.Time
target healthReporter
timer timer
unreachableCount int
}
func (s healthScheduler) ScheduledFor() time.Time {
return s.scheduledFor
}
// Reschedule, like the protocol described in scheduler, uses the current and
// proposed future health state to determine how and when a given subject is to
// be scheduled.
//
// If a subject has been at given moment marked as unhealthy, an exponential
// backoff scheme is applied to it. The reason for this backoff is to ensure
// that known-healthy targets can consume valuable request queuing resources
// first. Depending on the retrieval interval and number of consecutive
// unhealthy markings, the query of these unhealthy individuals may come before
// the healthy ones for a short time to help ensure expeditious retrieval.
// The inflection point that drops these to the back of the queue is beneficial
// to save resources in the long-run.
//
// If a subject is healthy, its next scheduling opportunity is set to
// earliest, for this ensures fair querying of all remaining healthy targets and
// removes bias in the ordering. In order for the anti-bias value to have any
// value, the earliest opportunity should be set to a value that is constant
// for a given batch of subjects who are to be scraped on a given interval.
func (s *healthScheduler) Reschedule(e time.Time, f TargetState) {
currentState := s.target.State()
// XXX: Handle metrics surrounding health.
switch currentState {
case UNKNOWN, UNREACHABLE:
switch f {
case ALIVE:
s.unreachableCount = 0
break
case UNREACHABLE:
s.unreachableCount++
break
}
case ALIVE:
switch f {
case UNREACHABLE:
s.unreachableCount++
}
}
if s.unreachableCount == 0 {
s.scheduledFor = e
} else {
backoff := MAXIMUM_BACKOFF_VALUE
exponential := time.Duration(math.Pow(DEFAULT_BACKOFF_VALUE, float64(s.unreachableCount))) * DEFAULT_BACKOFF_VALUE_UNIT
if backoff > exponential {
backoff = exponential
}
s.scheduledFor = s.timer.Now().Add(backoff)
}
}