func (t *Task) bdrPrecreateDatabase(client *consulapi.Client) (err error) {
/*
key := fmt.Sprintf(`rdpg/%s/cluster/service`, ClusterID)
kvp, _, err := kv.Get(key, nil)
if err != nil {
log.Error(fmt.Sprintf(`rdpg.RDPG<%s>#bdrGroupCreate() kv.Get() ! %s`, ClusterID, err))
return
}
v := ``
if kvp != nil {
v = string(kvp.Value)
}
if len(v) > 0 {
log.Trace(fmt.Sprintf(`rdpg.RDPG<%s>#bdrPrecreateDatabase()`, ClusterID))
return
}
*/
b := bdr.NewBDR(t.ClusterID, client)
re := regexp.MustCompile("[^A-Za-z0-9_]")
u1 := uuid.NewUUID().String()
u2 := uuid.NewUUID().String()
identifier := strings.ToLower(string(re.ReplaceAll([]byte(u1), []byte(""))))
dbpass := strings.ToLower(string(re.ReplaceAll([]byte(u2), []byte(""))))
i := &instances.Instance{
ClusterID: ClusterID,
Database: "d" + identifier,
User: "******" + identifier,
Pass: dbpass,
ClusterService: t.ClusterService,
}
// TODO: Keep the databases under rdpg schema, link to them in the
// cfsb.instances table so that we separate the concerns of CF and databases.
err = b.CreateUser(i.User, i.Pass)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#bdrPrecreateDatabases(%s) CreateUser(%s) ! %s", i.Database, i.User, err))
return err
}
err = b.CreateDatabase(i.Database, i.User)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#bdrPrecreateDatabases(%s) CreateDatabase(%s,%s) ! %s", i.Database, i.Database, i.User, err))
return err
}
p := pg.NewPG(`127.0.0.1`, pgPort, `rdpg`, `rdpg`, pgPass)
db, err := p.Connect()
if err != nil {
log.Error(fmt.Sprintf("tasks.Work() Failed connecting to %s err: %s", p.URI, err))
return err
}
defer db.Close()
sq := fmt.Sprintf(`INSERT INTO cfsb.instances (cluster_id,dbname, dbuser, dbpass, cluster_service) VALUES ('%s','%s','%s','%s','%s')`, ClusterID, i.Database, i.User, i.Pass, t.ClusterService)
log.Trace(fmt.Sprintf(`tasks.bdrPrecreateDatabase(%s) > %s`, i.Database, sq))
_, err = db.Query(sq)
if err != nil {
log.Error(fmt.Sprintf(`tasks.bdrPrecreateDatabase(%s) ! %s`, i.Database, err))
return err
}
err = b.CreateExtensions(i.Database, []string{`btree_gist`, `bdr`, `pg_stat_statements`, `uuid-ossp`, `hstore`, `pg_trgm`, `pgcrypto`})
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#bdrPrecreateDatabases(%s) CreateExtensions(%s,%s) ! %s", i.Database, i.Database, i.User, err))
return err
}
//Loop through and add any additional extensions specified in the rdpgd_service properties of the deployment manifest
if len(globals.UserExtensions) > 1 {
userExtensions := strings.Split(globals.UserExtensions, " ")
err = b.CreateExtensions(i.Database, userExtensions)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#bdrPrecreateDatabases(%s) CreateExtensions(%s,%s) Creating Extra User Extensions ! %s", i.Database, i.Database, i.User, err))
return err
}
}
err = b.CreateReplicationGroup(i.Database)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#bdrPrecreateDatabases(%s) CreateReplicationGroup() ! %s", i.Database, err))
return err
}
sq = fmt.Sprintf(`UPDATE cfsb.instances SET effective_at=CURRENT_TIMESTAMP WHERE dbname='%s'`, i.Database)
log.Trace(fmt.Sprintf(`tasks.bdrPrecreateDatabase(%s) > %s`, i.Database, sq))
_, err = db.Query(sq)
if err != nil {
log.Error(fmt.Sprintf(`tasks.bdrPrecreateDatabase(%s) ! %s`, i.Database, err))
return err
}
// Tell the management cluster about the newly available database.
// TODO: This can be in a function.
catalog := client.Catalog()
svcs, _, err := catalog.Service("rdpgmc", "", nil)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#bdrPrecreateDatabase(%s) catalog.Service() ! %s", i.Database, err))
return err
}
if len(svcs) == 0 {
log.Error(fmt.Sprintf("tasks.Task#bdrPrecreateDatabase(%s) ! No services found, no known nodes?!", i.Database))
return err
}
body, err := json.Marshal(i)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#bdrPrecreateDatabase(%s) json.Marchal(i) ! %s", i.Database, err))
return err
}
url := fmt.Sprintf("http://%s:%s/%s", svcs[0].Address, os.Getenv("RDPGD_ADMIN_PORT"), `databases/register`)
req, err := http.NewRequest("POST", url, bytes.NewBuffer([]byte(body)))
if err != nil {
log.Error(fmt.Sprintf(`tasks.Task#bdrPrecreateDatabase(%s) http.NewRequest() POST %s ! %s`, i.Database, url, err))
return err
}
log.Trace(fmt.Sprintf(`tasks.Task#bdrPrecreateDatabase(%s) POST %s`, i.Database, url))
req.SetBasicAuth(os.Getenv("RDPGD_ADMIN_USER"), os.Getenv("RDPGD_ADMIN_PASS"))
httpClient := &http.Client{}
resp, err := httpClient.Do(req)
if err != nil {
log.Error(fmt.Sprintf(`tasks.Task#bdrPrecreateDatabase(%s) httpClient.Do() %s ! %s`, i.Database, url, err))
return err
}
resp.Body.Close()
return
}
// Precreate database functionality, note that the database creation lock is
// expected to be held when this is called as this must be done in sequence.
func (t *Task) precreateDatabase(workRole string, client *consulapi.Client) (err error) {
log.Trace(fmt.Sprintf("tasks.Task#precreateDatabase()..."))
b := bdr.NewBDR(t.ClusterID, client)
re := regexp.MustCompile("[^A-Za-z0-9_]")
u1 := uuid.NewUUID().String()
u2 := uuid.NewUUID().String()
identifier := strings.ToLower(string(re.ReplaceAll([]byte(u1), []byte(""))))
dbpass := strings.ToLower(string(re.ReplaceAll([]byte(u2), []byte(""))))
i := &instances.Instance{
ClusterID: ClusterID,
Database: "d" + identifier,
User: "******" + identifier,
Pass: dbpass,
}
// TODO: Keep the databases under rdpg schema, link to them in the
// cfsb.instances table so that we separate the concerns of CF and databases.
err = b.CreateUser(i.User, i.Pass)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#PrecreateDatabases(%s) CreateUser(%s) ! %s", i.Database, i.User, err))
return err
}
err = b.CreateDatabase(i.Database, i.User)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#PrecreateDatabases(%s) CreateDatabase(%s,%s) ! %s", i.Database, i.Database, i.User, err))
return err
}
p := pg.NewPG(`127.0.0.1`, pbPort, `rdpg`, `rdpg`, pgPass)
db, err := p.Connect()
if err != nil {
log.Error(fmt.Sprintf("tasks.Work() Failed connecting to %s err: %s", p.URI, err))
return err
}
defer db.Close()
sq := fmt.Sprintf(`INSERT INTO cfsb.instances (cluster_id,dbname, dbuser, dbpass) VALUES ('%s','%s','%s','%s')`, ClusterID, i.Database, i.User, i.Pass)
log.Trace(fmt.Sprintf(`tasks.precreateDatabase(%s) > %s`, i.Database, sq))
_, err = db.Query(sq)
if err != nil {
log.Error(fmt.Sprintf(`tasks.precreateDatabase(%s) ! %s`, i.Database, err))
return err
}
err = b.CreateExtensions(i.Database, []string{`btree_gist`, `bdr`})
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#PrecreateDatabases(%s) CreateExtensions(%s,%s) ! %s", i.Database, i.Database, i.User, err))
return err
}
err = b.CreateReplicationGroup(i.Database)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#PrecreateDatabases(%s) CreateReplicationGroup() ! %s", i.Database, err))
return err
}
sq = fmt.Sprintf(`UPDATE cfsb.instances SET effective_at=CURRENT_TIMESTAMP WHERE dbname='%s'`, i.Database)
log.Trace(fmt.Sprintf(`tasks.precreateDatabase(%s) > %s`, i.Database, sq))
_, err = db.Query(sq)
if err != nil {
log.Error(fmt.Sprintf(`tasks.precreateDatabase(%s) ! %s`, i.Database, err))
return err
}
// Tell the management cluster about the newly available database.
// TODO: This can be in a function.
catalog := client.Catalog()
svcs, _, err := catalog.Service("rdpgmc", "", nil)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#precreateDatabase(%s) catalog.Service() ! %s", i.Database, err))
return err
}
if len(svcs) == 0 {
log.Error(fmt.Sprintf("tasks.Task#precreateDatabase(%s) ! No services found, no known nodes?!", i.Database))
return err
}
body, err := json.Marshal(i)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#precreateDatabase(%s) json.Marchal(i) ! %s", i.Database, err))
return err
}
url := fmt.Sprintf("http://%s:%s/%s", svcs[0].Address, os.Getenv("RDPGD_ADMIN_PORT"), `databases/register`)
req, err := http.NewRequest("POST", url, bytes.NewBuffer([]byte(body)))
log.Trace(fmt.Sprintf(`tasks.Task#precreateDatabase(%s) POST %s`, i.Database, url))
// req.Header.Set("Content-Type", "application/json")
// TODO: Retrieve from configuration in database.
req.SetBasicAuth(os.Getenv("RDPGD_ADMIN_USER"), os.Getenv("RDPGD_ADMIN_PASS"))
httpClient := &http.Client{}
resp, err := httpClient.Do(req)
if err != nil {
log.Error(fmt.Sprintf(`tasks.Task#precreateDatabase(%s) httpClient.Do() %s ! %s`, i.Database, url, err))
return err
}
resp.Body.Close()
return
}
// TODO: This should be remove database
func (t *Task) RemoveDatabase(workRole string) (err error) {
// For now we assume data is simply the database name.
key := fmt.Sprintf("rdpg/%s/work/databases/remove", os.Getenv(`RDPGD_CLUSTER`))
client, _ := consulapi.NewClient(consulapi.DefaultConfig())
lock, err := client.LockKey(key)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#RemoveDatabase() Error aquiring lock ! %s", err))
return
}
leaderCh, err := lock.Lock(nil)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#RemoveDatabase() Error aquiring lock ! %s", err))
return
}
if leaderCh == nil {
log.Trace(fmt.Sprintf("tasks.Task#RemoveDatabase() > Not Leader."))
return
}
log.Trace(fmt.Sprintf("tasks.Task#RemoveDatabase() > Leader."))
p := pg.NewPG(`127.0.0.1`, pbPort, `rdpg`, `rdpg`, pgPass)
db, err := p.Connect()
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#RemoveDatabase() p.Connect() ! %s", err))
return
}
ids := []string{}
sq := fmt.Sprintf(`SELECT instance_id from cfsb.instances WHERE ineffective_at IS NOT NULL AND ineffective_at < CURRENT_TIMESTAMP AND decommissioned_at IS NULL`)
err = db.Select(&ids, sq)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#RemoveDatabase() Querying for Databases to Cleanup ! %s", err))
}
db.Close()
for _, id := range ids {
db, err := p.Connect()
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#RemoveDatabase() p.Connect() ! %s", err))
return err
}
uri := "postgres://"
b := bdr.NewBDR(uri, client)
i, err := instances.FindByInstanceID(id)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#RemoveDatabase(%s) FindingInstance(%s) ! %s", i.Database, i.InstanceID, err))
db.Close()
continue
}
err = b.DisableDatabase(i.Database)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#RemoveDatabase() DisableDatabase(%s) for %s ! %s", i.Database, i.InstanceID, err))
db.Close()
continue
}
err = b.BackupDatabase(i.Database)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#RemoveDatabase() BackupDatabase(%s) ! %s", i.Database, err))
db.Close()
continue
}
// Question, How do we "stop" the replication group before dropping the database?
err = b.DropDatabase(i.Database)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#RemoveDatabase() DropDatabase(%s) for %s ! %s", i.Database, i.InstanceID, err))
db.Close()
continue
}
err = b.DropUser(i.User)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#RemoveDatabase() DropUser(%s) for %s ! %s", i.User, i.InstanceID, err))
db.Close()
continue
}
err = b.DropDatabase(i.Database)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#RemoveDatabase() DropDatabase(%s) for %s ! %s", i.Database, i.InstanceID, err))
db.Close()
continue
}
}
db.Close()
return
}
//DecommissionDatabase - Remove targeted database specified in Data
func (t *Task) DecommissionDatabase() (err error) {
log.Trace(fmt.Sprintf(`tasks.DecommissionDatabase(%s)...`, t.Data))
i, err := instances.FindByDatabase(t.Data)
if err != nil {
log.Error(fmt.Sprintf("tasks.DecommissionDatabase(%s) instances.FindByDatabase() ! %s", i.Database, err))
return err
}
//TODO: Check if i == nil; i.e. if database doesn't exist
ips, err := i.ClusterIPs()
if err != nil {
log.Error(fmt.Sprintf(`tasks.Task#DecommissionDatabase(%s) i.ClusterIPs() ! %s`, i.Database, err))
return err
}
if len(ips) == 0 {
log.Error(fmt.Sprintf("tasks.Task#DecommissionDatabase(%s) ! No service cluster nodes found in Consul?!", i.Database))
return
}
p := pg.NewPG(`127.0.0.1`, pbPort, `rdpg`, `rdpg`, pgPass)
db, err := p.Connect()
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#DecommissionDatabase(%s) p.Connect(%s) ! %s", t.Data, p.URI, err))
return err
}
defer db.Close()
switch globals.ServiceRole {
case "manager":
path := fmt.Sprintf(`databases/decommission/%s`, t.Data)
url := fmt.Sprintf("http://%s:%s/%s", ips[0], os.Getenv("RDPGD_ADMIN_PORT"), path)
req, err := http.NewRequest("DELETE", url, bytes.NewBuffer([]byte("{}")))
log.Trace(fmt.Sprintf(`tasks.Task#Decommission() > DELETE %s`, url))
//req.Header.Set("Content-Type", "application/json")
// TODO: Retrieve from configuration in database.
req.SetBasicAuth(os.Getenv("RDPGD_ADMIN_USER"), os.Getenv("RDPGD_ADMIN_PASS"))
httpClient := &http.Client{}
_, err = httpClient.Do(req)
if err != nil {
log.Error(fmt.Sprintf(`tasks.Task#DecommissionDatabase(%s) httpClient.Do() %s ! %s`, i.Database, url, err))
return err
}
// TODO: Is there anything we want to do on successful request?
case "service":
// In here we must do everything necessary to physically delete and clean up
// the database from all service cluster nodes.
if err = t.BackupDatabase(); err != nil {
log.Error(fmt.Sprintf(`tasks.Task#DecommissionDatabase(%s) t.BackupDatabase(%s) ! %s`, i.Database, err))
} else {
for _, ip := range ips { // Schedule pgbouncer reconfigure on each cluster node.
newTask := Task{ClusterID: ClusterID, Node: ip, Role: "all", Action: "Reconfigure", Data: "pgbouncer", NodeType: "any"}
err = newTask.Enqueue()
if err != nil {
log.Error(fmt.Sprintf(`tasks.Task#DecommissionDatabase(%s) Reconfigure PGBouncer! %s`, i.Database, err))
}
}
log.Trace(fmt.Sprintf(`tasks.DecommissionDatabase(%s) TODO: Here is where we finally decommission on the service cluster...`, i.Database))
client, err := consulapi.NewClient(consulapi.DefaultConfig())
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#DecommissionDatabase(%s) consulapi.NewClient() ! %s", i.Database, err))
return err
}
// Lock Database Deletion via Consul Lock
key := fmt.Sprintf(`rdpg/%s/database/existance/lock`, t.ClusterID)
lo := &consulapi.LockOptions{
Key: key,
SessionName: fmt.Sprintf(`rdpg/%s/databases/existance`, t.ClusterID),
}
log.Trace(fmt.Sprintf(`tasks.Task<%s>#DecommissionDatabase() Attempting to acquire database existance lock %s...`, t.ClusterID, key))
databaseCreateLock, err := client.LockOpts(lo)
if err != nil {
log.Error(fmt.Sprintf(`tasks.Task<%s>#DecommissionDatabase() LockKey() database/existance Lock Key %s ! %s`, t.ClusterID, key, err))
return err
}
databaseCreateLockCh, err := databaseCreateLock.Lock(nil)
if err != nil {
log.Error(fmt.Sprintf(`tasks.Task<%s>#DecommissionDatabase() Lock() database/existance lock %s ! %s`, t.ClusterID, key, err))
return err
}
if databaseCreateLockCh == nil {
err := fmt.Errorf(`tasks.Task<%s>#DecommissionDatabase() database/existance Lock not aquired, halting Decommission!!!`, t.ClusterID)
log.Error(err.Error())
return err
}
defer databaseCreateLock.Unlock()
p := pg.NewPG(`127.0.0.1`, pbPort, `rdpg`, `rdpg`, pgPass)
db, err := p.Connect()
if err != nil {
log.Error(fmt.Sprintf("instances.Decommission() p.Connect(%s) ! %s", p.URI, err))
return err
}
defer db.Close()
sq := fmt.Sprintf(`DELETE FROM tasks.tasks WHERE action='BackupDatabase' AND data='%s'`, i.Database)
log.Trace(fmt.Sprintf(`tasks.Task#DecommissionDatabase(%s) SQL > %s`, i.Database, sq))
_, err = db.Exec(sq)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#DecommissionDatabase(%s) ! %s", i.Database, err))
}
sq = fmt.Sprintf(`UPDATE tasks.schedules SET enabled = false WHERE action='BackupDatabase' AND data='%s'`, i.Database)
log.Trace(fmt.Sprintf(`tasks.Task#DecommissionDatabase(%s) SQL > %s`, i.Database, sq))
_, err = db.Exec(sq)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#DecommissionDatabase(%s) ! %s", i.Database, err))
}
if t.ClusterService == "pgbdr" {
b := bdr.NewBDR(ClusterID, client)
b.DropDatabase(i.Database)
dbuser := ""
sq = fmt.Sprintf(`SELECT dbuser FROM cfsb.instances WHERE dbname='%s' LIMIT 1`, i.Database)
log.Trace(fmt.Sprintf(`tasks.Task#DecommissionDatabase(%s) SQL > %s`, i.Database, sq))
err = db.Get(&dbuser, sq)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#DecommissionDatabase(%s) ! %s", i.Database, err))
}
b.DropUser(dbuser)
sq = fmt.Sprintf(`UPDATE cfsb.instances SET decommissioned_at=CURRENT_TIMESTAMP WHERE dbname='%s'`, i.Database)
log.Trace(fmt.Sprintf(`tasks.Task#DecommissionDatabase(%s) SQL > %s`, i.Database, sq))
_, err = db.Exec(sq)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#DecommissionDatabase(%s) ! %s", i.Database, err))
}
} else {
p.DisableDatabase(i.Database)
p.DropDatabase(i.Database)
dbuser := ""
sq = fmt.Sprintf(`SELECT dbuser FROM cfsb.instances WHERE dbname='%s' LIMIT 1`, i.Database)
log.Trace(fmt.Sprintf(`tasks.Task#DecommissionDatabase(%s) SQL > %s`, i.Database, sq))
err = db.Get(&dbuser, sq)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#DecommissionDatabase(%s) ! %s", i.Database, err))
}
p.DropUser(dbuser)
sq = fmt.Sprintf(`UPDATE cfsb.instances SET decommissioned_at=CURRENT_TIMESTAMP WHERE dbname='%s'`, i.Database)
log.Trace(fmt.Sprintf(`tasks.Task#DecommissionDatabase(%s) SQL > %s`, i.Database, sq))
_, err = db.Exec(sq)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#DecommissionDatabase(%s) ! %s", i.Database, err))
}
}
// Notify management cluster that the instance has been decommissioned
// Find management cluster API address
catalog := client.Catalog()
svcs, _, err := catalog.Service(`rdpgmc`, "", nil)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#DecommissionDatabase(%s) consulapi.Client.Catalog() ! %s", i.Database, err))
return err
}
if len(svcs) == 0 {
log.Error(fmt.Sprintf("tasks.Task#DecommissionDatabase(%s) ! No services found, no known nodes?!", i.Database))
return err
}
mgtAPIIPAddress := svcs[0].Address
// Query the database for the decommissioned_at timestamp set
timestamp := ""
sq = fmt.Sprintf(`SELECT decommissioned_at::text FROM cfsb.instances WHERE dbname='%s' LIMIT 1;`, i.Database)
db.Get(×tamp, sq)
type decomm struct {
Database string `json:"database"`
Timestamp string `json:"timestamp"`
}
dc := decomm{Database: i.Database, Timestamp: timestamp}
// Tell the management cluster (via admin api) about the timestamp.
url := fmt.Sprintf("http://%s:%s/%s", mgtAPIIPAddress, os.Getenv("RDPGD_ADMIN_PORT"), `databases/decommissioned`)
body, err := json.Marshal(dc)
if err != nil {
log.Error(fmt.Sprintf("tasks.Task#DecommissionDatabase(%s) json.Marchal(i) ! %s", i.Database, err))
return err
}
req, err := http.NewRequest("PUT", url, bytes.NewBuffer([]byte(body)))
log.Trace(fmt.Sprintf(`tasks.Task#DecommissionDatabase(%s) PUT %s body: %s`, i.Database, url, body))
req.SetBasicAuth(os.Getenv("RDPGD_ADMIN_USER"), os.Getenv("RDPGD_ADMIN_PASS"))
httpClient := &http.Client{}
resp, err := httpClient.Do(req)
if err != nil {
log.Error(fmt.Sprintf(`tasks.Task#DecommissionDatabase(%s) httpClient.Do() PUT %s ! %s`, i.Database, url, err))
return err
}
resp.Body.Close()
}
return nil
default:
log.Error(fmt.Sprintf(`tasks.Task#DecommissionDatabase(%s) ! Unknown work role: '%s' -> BUG!!!`, i.Database, globals.ServiceRole))
return nil
}
return
}
