forked from TrueCloudLab/frostfs-node
[#723] netmap: Drop already bootstraped check
Because of this check, under certain conditions, the node could be removed from the network map, although the node was functioning normally. Signed-off-by: Dmitrii Stepanov <d.stepanov@yadro.com>
This commit is contained in:
parent
eb5248621a
commit
e0f0b93b5e
2 changed files with 7 additions and 72 deletions
|
@ -345,8 +345,7 @@ type internals struct {
|
|||
apiVersion version.Version
|
||||
healthStatus *atomic.Int32
|
||||
// is node under maintenance
|
||||
isMaintenance atomic.Bool
|
||||
alreadyBootstraped bool
|
||||
isMaintenance atomic.Bool
|
||||
}
|
||||
|
||||
// starts node's maintenance.
|
||||
|
|
|
@ -220,10 +220,6 @@ func bootstrapNode(c *cfg) {
|
|||
c.log.Info(logs.FrostFSNodeNodeIsUnderMaintenanceSkipInitialBootstrap)
|
||||
return
|
||||
}
|
||||
if c.alreadyBootstraped {
|
||||
c.log.Info(logs.NetmapNodeAlreadyInCandidateListOnlineSkipInitialBootstrap)
|
||||
return
|
||||
}
|
||||
err := c.bootstrap()
|
||||
fatalOnErrDetails("bootstrap error", err)
|
||||
}
|
||||
|
@ -256,7 +252,7 @@ func initNetmapState(c *cfg) {
|
|||
fatalOnErrDetails("could not initialize current epoch number", err)
|
||||
|
||||
var ni *netmapSDK.NodeInfo
|
||||
ni, c.alreadyBootstraped, err = c.netmapInitLocalNodeState(epoch)
|
||||
ni, err = c.netmapInitLocalNodeState(epoch)
|
||||
fatalOnErrDetails("could not init network state", err)
|
||||
|
||||
stateWord := nodeState(ni)
|
||||
|
@ -275,64 +271,6 @@ func initNetmapState(c *cfg) {
|
|||
c.handleLocalNodeInfo(ni)
|
||||
}
|
||||
|
||||
func needsUpdate(local, remote *netmapSDK.NodeInfo) bool {
|
||||
return bytes.Equal(local.PublicKey(), remote.PublicKey()) && equalEndpoints(local, remote) && equalAttributes(local, remote)
|
||||
}
|
||||
|
||||
func equalAttributes(local, remote *netmapSDK.NodeInfo) bool {
|
||||
asA := make(map[string]string)
|
||||
local.IterateAttributes(func(k, v string) {
|
||||
asA[k] = v
|
||||
})
|
||||
|
||||
allMatched := true
|
||||
count := 0
|
||||
remote.IterateAttributes(func(k, vb string) {
|
||||
// IR adds new attributes derived from the locode, they should be skipped.
|
||||
if isLocodeAttribute(k) {
|
||||
return
|
||||
}
|
||||
if va, ok := asA[k]; !ok || va != vb {
|
||||
allMatched = false
|
||||
return
|
||||
}
|
||||
count++
|
||||
})
|
||||
return allMatched && count == len(asA)
|
||||
}
|
||||
|
||||
func isLocodeAttribute(k string) bool {
|
||||
// See https://git.frostfs.info/TrueCloudLab/frostfs-api/src/branch/master/netmap/types.proto#L171
|
||||
switch k {
|
||||
case "Continent", "Country", "CountryCode", "Location", "SubDiv", "SubDivCode":
|
||||
return true
|
||||
default:
|
||||
return false
|
||||
}
|
||||
}
|
||||
|
||||
func equalEndpoints(a, b *netmapSDK.NodeInfo) bool {
|
||||
var esA, esB []string
|
||||
a.IterateNetworkEndpoints(func(e string) bool {
|
||||
esA = append(esA, e)
|
||||
return false
|
||||
})
|
||||
b.IterateNetworkEndpoints(func(e string) bool {
|
||||
esB = append(esB, e)
|
||||
return false
|
||||
})
|
||||
|
||||
if len(esA) != len(esB) {
|
||||
return false
|
||||
}
|
||||
for i := range esA {
|
||||
if esA[i] != esB[i] {
|
||||
return false
|
||||
}
|
||||
}
|
||||
return true
|
||||
}
|
||||
|
||||
func nodeState(ni *netmapSDK.NodeInfo) string {
|
||||
if ni != nil {
|
||||
switch {
|
||||
|
@ -347,29 +285,27 @@ func nodeState(ni *netmapSDK.NodeInfo) string {
|
|||
return "undefined"
|
||||
}
|
||||
|
||||
func (c *cfg) netmapInitLocalNodeState(epoch uint64) (*netmapSDK.NodeInfo, bool, error) {
|
||||
func (c *cfg) netmapInitLocalNodeState(epoch uint64) (*netmapSDK.NodeInfo, error) {
|
||||
nmNodes, err := c.cfgNetmap.wrapper.GetCandidates()
|
||||
if err != nil {
|
||||
return nil, false, err
|
||||
return nil, err
|
||||
}
|
||||
|
||||
var candidate *netmapSDK.NodeInfo
|
||||
alreadyBootstraped := false
|
||||
for i := range nmNodes {
|
||||
if bytes.Equal(nmNodes[i].PublicKey(), c.binPublicKey) {
|
||||
candidate = &nmNodes[i]
|
||||
alreadyBootstraped = candidate.IsOnline() && needsUpdate(&c.cfgNodeInfo.localInfo, candidate)
|
||||
break
|
||||
}
|
||||
}
|
||||
|
||||
node, err := c.netmapLocalNodeState(epoch)
|
||||
if err != nil {
|
||||
return nil, false, err
|
||||
return nil, err
|
||||
}
|
||||
|
||||
if candidate == nil {
|
||||
return node, false, nil
|
||||
return node, nil
|
||||
}
|
||||
|
||||
nmState := nodeState(node)
|
||||
|
@ -381,7 +317,7 @@ func (c *cfg) netmapInitLocalNodeState(epoch uint64) (*netmapSDK.NodeInfo, bool,
|
|||
zap.String("netmap", nmState),
|
||||
zap.String("candidate", candidateState))
|
||||
}
|
||||
return candidate, alreadyBootstraped, nil
|
||||
return candidate, nil
|
||||
}
|
||||
|
||||
func (c *cfg) netmapLocalNodeState(epoch uint64) (*netmapSDK.NodeInfo, error) {
|
||||
|
|
Loading…
Reference in a new issue