README
¶
Coordinator
Support both zookeeper and etcd
Documentation
¶
Overview ¶
Package fakezk is a pretty complete mock implementation of a Zookeper connection (see go/zk/zk.Conn). All operations work as expected with the exceptions of zk.Conn.ACL and zk.Conn.SetACL. zk.Conn.SetACL will succeed, but it is a noop (and the ACLs won't be respected). zk.Conn.ACL will panic. It is OK to access the connection from multiple goroutines, but the locking is very naive (every operation locks the whole connection).
Index ¶
- Constants
- Variables
- func ChildrenRecursive(zconn Conn, zkPath string) ([]string, error)
- func CreateOrUpdate(zconn Conn, zkPath, value string, flags int, aclv []zk.ACL, recursive bool) (pathCreated string, err error)
- func CreatePidNode(zconn Conn, zkPath string, contents string, done chan struct{}) error
- func CreateRecursive(zconn Conn, zkPath, value string, flags int, aclv []zk.ACL) (pathCreated string, err error)
- func DefaultACLs() []zk.ACL
- func DefaultDirACLs() []zk.ACL
- func DefaultFileACLs() []zk.ACL
- func DeleteRecursive(zconn Conn, zkPath string, version int) error
- func HasWildcard(path string) bool
- func IsDirectory(aclv []zk.ACL) bool
- func NodeExists(zconn Conn, zkPath string) (bool, error)
- func ObtainQueueLock(zconn Conn, zkPath string, wait time.Duration, interrupted chan struct{}) error
- func ResolveWildcards(zconn Conn, zkPaths []string) ([]string, error)
- func ZkErrorEqual(a, b error) bool
- func ZkEventOk(e zk.Event) bool
- type Conn
- type ElectorTask
- type MyZkConn
- type PooledEtcdClient
- type ZElector
- type ZLocker
- type ZkError
Constants ¶
const ( // PERM_DIRECTORY are default permissions for a node. PERM_DIRECTORY = zk.PermAdmin | zk.PermCreate | zk.PermDelete | zk.PermRead | zk.PermWrite // PERM_FILE allows a zk node to emulate file behavior by disallowing child nodes. PERM_FILE = zk.PermAdmin | zk.PermRead | zk.PermWrite MagicPrefix = "zk" )
const MAX_TTL = 365 * 24 * 60 * 60
Variables ¶
var ( // This error is returned by functions that wait for a result // when they are interrupted. ErrInterrupted = errors.New("zkutil: obtaining lock was interrupted") // This error is returned by functions that wait for a result // when the timeout value is reached. ErrTimeout = errors.New("zkutil: obtaining lock timed out") )
Functions ¶
func CreateOrUpdate ¶
func CreatePidNode ¶
Close the release channel when you want to clean up nicely.
func CreateRecursive ¶
func CreateRecursive(zconn Conn, zkPath, value string, flags int, aclv []zk.ACL) (pathCreated string, err error)
Create a path and any pieces required, think mkdir -p. Intermediate znodes are always created empty.
func DefaultACLs ¶
func DefaultDirACLs ¶
func DefaultFileACLs ¶
func HasWildcard ¶
func IsDirectory ¶
IsDirectory returns if this node should be treated as a directory.
func ObtainQueueLock ¶
func ObtainQueueLock(zconn Conn, zkPath string, wait time.Duration, interrupted chan struct{}) error
The lexically lowest node is the lock holder - verify that this path holds the lock. Call this queue-lock because the semantics are a hybrid. Normal zk locks make assumptions about sequential numbering that don't hold when the data in a lock is modified. if the provided 'interrupted' chan is closed, we'll just stop waiting and return an interruption error
func ResolveWildcards ¶
resolve paths like: /zk/nyc/vt/tablets/*/action /zk/global/vt/keyspaces/*/shards/*/action /zk/*/vt/tablets/*/action into real existing paths
If you send paths that don't contain any wildcard and don't exist, this function will return an empty array.
func ZkErrorEqual ¶
Types ¶
type Conn ¶
type Conn interface {
Get(path string) (data []byte, stat zk.Stat, err error)
GetW(path string) (data []byte, stat zk.Stat, watch <-chan zk.Event, err error)
Children(path string) (children []string, stat zk.Stat, err error)
ChildrenW(path string) (children []string, stat zk.Stat, watch <-chan zk.Event, err error)
Exists(path string) (exist bool, stat zk.Stat, err error)
ExistsW(path string) (exist bool, stat zk.Stat, watch <-chan zk.Event, err error)
Create(path string, value []byte, flags int32, aclv []zk.ACL) (pathCreated string, err error)
Set(path string, value []byte, version int32) (stat zk.Stat, err error)
Delete(path string, version int32) (err error)
Close()
GetACL(path string) ([]zk.ACL, zk.Stat, error)
SetACL(path string, aclv []zk.ACL, version int32) (zk.Stat, error)
Seq2Str(seq int64) string
}
This interface is really close to the zookeeper connection interface. It uses the Stat interface defined here instead of the zookeeper.Stat structure for stats. Everything else is the same as in zookeeper. So refer to the zookeeper docs for the conventions used here (for instance, using -1 as version to specify any version)
func ConnectToZk ¶
func NewConn ¶
func NewConn() Conn
NewConn returns a fake zk.Conn implementation. Data is stored in memory, and there's a global connection lock for concurrent access.
func NewConnFromFile ¶
NewConnFromFile returns a fake zk.Conn implementation, that is seeded with the json data extracted from the input file.
func NewEtcdConn ¶
type ElectorTask ¶
type ElectorTask interface {
Run() error
Stop()
// Return true if interrupted, false if it died of natural causes.
// An interrupted task indicates that the election should stop.
Interrupted() bool
}
ElectorTask is the interface for a task that runs essentially forever or until something bad happens. If a task must be stopped, it should be handled promptly - no second notification will be sent.
type PooledEtcdClient ¶
type PooledEtcdClient struct {
// contains filtered or unexported fields
}
func (*PooledEtcdClient) Close ¶
func (c *PooledEtcdClient) Close()
type ZElector ¶
type ZElector struct {
// contains filtered or unexported fields
}
ZElector stores basic state for running an election.
func CreateElection ¶
CreateElection returns an initialized elector. An election is really a cycle of events. You are flip-flopping between leader and candidate. It's better to think of this as a stream of events that one needs to react to.
func (ZElector) LockWithTimeout ¶
LockWithTimeout returns nil when the lock is acquired. A lock is held if the file exists and you are the creator. Setting the wait to zero makes this a nonblocking lock check.
FIXME(msolo) Disallow non-super users from removing the lock?
func (*ZElector) RunTask ¶
func (ze *ZElector) RunTask(task ElectorTask) error
RunTask returns nil when the underlyingtask ends or the error it generated.
type ZLocker ¶
type ZLocker interface {
Lock(desc string) error
LockWithTimeout(wait time.Duration, desc string) error
Unlock() error
Interrupt()
}
ZLocker is an interface for a lock that can fail.
func CreateMutex ¶
CreateMutex initializes an unaquired mutex. A mutex is released only by Unlock. You can clean up a mutex with delete, but you should be careful doing so.
Source Files