forked from colinmarc/hdfs
/
datanode_failover.go
71 lines (57 loc) · 1.53 KB
/
datanode_failover.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
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
package rpc
import (
"sync"
"time"
)
// datanodeFailures is a global map of address to the last recorded failure
var datanodeFailures = make(map[string]time.Time)
var datanodeFailuresLock sync.Mutex
// a datanodeFailover provides some common code for trying multiple datanodes
// in the context of a single operation on a single block.
type datanodeFailover struct {
datanodes []string
currentDatanode string
err error
}
func newDatanodeFailover(datanodes []string) *datanodeFailover {
return &datanodeFailover{
datanodes: datanodes,
currentDatanode: "",
err: nil,
}
}
func (df *datanodeFailover) recordFailure(err error) {
datanodeFailuresLock.Lock()
defer datanodeFailuresLock.Unlock()
datanodeFailures[df.currentDatanode] = time.Now()
df.err = err
}
func (df *datanodeFailover) next() string {
if df.numRemaining() == 0 {
return ""
}
var picked = -1
var oldestFailure time.Time
for i, address := range df.datanodes {
datanodeFailuresLock.Lock()
failedAt, hasFailed := datanodeFailures[address]
datanodeFailuresLock.Unlock()
if !hasFailed {
picked = i
break
} else if oldestFailure.IsZero() || failedAt.Before(oldestFailure) {
picked = i
oldestFailure = failedAt
}
}
address := df.datanodes[picked]
df.datanodes = append(df.datanodes[:picked], df.datanodes[picked+1:]...)
df.currentDatanode = address
return address
}
func (df *datanodeFailover) numRemaining() int {
return len(df.datanodes)
}
func (df *datanodeFailover) lastError() error {
return df.err
}