diff --git a/asynq.go b/asynq.go
index ba463b6..1ffcc41 100644
--- a/asynq.go
+++ b/asynq.go
@@ -9,32 +9,18 @@ TODOs:
 */
 
 import (
-	"encoding/json"
-	"fmt"
-	"math"
-	"math/rand"
-	"strconv"
 	"sync"
 	"time"
 
 	"github.com/go-redis/redis/v7"
 )
 
-// Redis keys
-const (
-	queuePrefix = "asynq:queues:"   // LIST
-	allQueues   = "asynq:queues"    // SET
-	scheduled   = "asynq:scheduled" // ZSET
-	retry       = "asynq:retry"     // ZSET
-	dead        = "asynq:dead"      // ZSET
-)
-
 // Max retry count by default
 const defaultMaxRetry = 25
 
 // Client is an interface for scheduling tasks.
 type Client struct {
-	rdb *redis.Client
+	rdb *rdb
 }
 
 // Task represents a task to be performed.
@@ -76,8 +62,8 @@ type RedisOpt struct {
 
 // NewClient creates and returns a new client.
 func NewClient(opt *RedisOpt) *Client {
-	rdb := redis.NewClient(&redis.Options{Addr: opt.Addr, Password: opt.Password})
-	return &Client{rdb: rdb}
+	client := redis.NewClient(&redis.Options{Addr: opt.Addr, Password: opt.Password})
+	return &Client{rdb: newRDB(client)}
 }
 
 // Process enqueues the task to be performed at a given time.
@@ -94,17 +80,15 @@ func (c *Client) Process(task *Task, executeAt time.Time) error {
 // enqueue pushes a given task to the specified queue.
 func (c *Client) enqueue(msg *taskMessage, executeAt time.Time) error {
 	if time.Now().After(executeAt) {
-		return push(c.rdb, msg)
+		return c.rdb.push(msg)
 	}
-	return zadd(c.rdb, scheduled, float64(executeAt.Unix()), msg)
+	return c.rdb.zadd(scheduled, float64(executeAt.Unix()), msg)
 }
 
 //-------------------- Launcher --------------------
 
 // Launcher starts the manager and poller.
 type Launcher struct {
-	rdb *redis.Client
-
 	// running indicates whether manager and poller are both running.
 	running bool
 	mu      sync.Mutex
@@ -116,16 +100,11 @@ type Launcher struct {
 
 // NewLauncher creates and returns a new Launcher.
 func NewLauncher(poolSize int, opt *RedisOpt) *Launcher {
-	rdb := redis.NewClient(&redis.Options{Addr: opt.Addr, Password: opt.Password})
-	poller := &poller{
-		rdb:         rdb,
-		done:        make(chan struct{}),
-		avgInterval: 5 * time.Second,
-		zsets:       []string{scheduled, retry},
-	}
+	client := redis.NewClient(&redis.Options{Addr: opt.Addr, Password: opt.Password})
+	rdb := newRDB(client)
+	poller := newPoller(rdb, 5*time.Second, []string{scheduled, retry})
 	manager := newManager(rdb, poolSize, nil)
 	return &Launcher{
-		rdb:     rdb,
 		poller:  poller,
 		manager: manager,
 	}
@@ -160,60 +139,3 @@ func (l *Launcher) Stop() {
 	l.poller.terminate()
 	l.manager.terminate()
 }
-
-// push pushes the task to the specified queue to get picked up by a worker.
-func push(rdb *redis.Client, msg *taskMessage) error {
-	bytes, err := json.Marshal(msg)
-	if err != nil {
-		return fmt.Errorf("could not encode task into JSON: %v", err)
-	}
-	qname := queuePrefix + msg.Queue
-	err = rdb.SAdd(allQueues, qname).Err()
-	if err != nil {
-		return fmt.Errorf("could not execute command SADD %q %q: %v",
-			allQueues, qname, err)
-	}
-	return rdb.RPush(qname, string(bytes)).Err()
-}
-
-// zadd serializes the given message and adds to the specified sorted set.
-func zadd(rdb *redis.Client, zset string, zscore float64, msg *taskMessage) error {
-	bytes, err := json.Marshal(msg)
-	if err != nil {
-		return fmt.Errorf("could not encode task into JSON: %v", err)
-	}
-	return rdb.ZAdd(zset, &redis.Z{Member: string(bytes), Score: zscore}).Err()
-}
-
-const maxDeadTask = 100
-const deadExpirationInDays = 90
-
-// kill sends the task to "dead" sorted set. It also trim the sorted set by
-// timestamp and set size.
-func kill(rdb *redis.Client, msg *taskMessage) error {
-	bytes, err := json.Marshal(msg)
-	if err != nil {
-		return fmt.Errorf("could not encode task into JSON: %v", err)
-	}
-	now := time.Now()
-	pipe := rdb.Pipeline()
-	pipe.ZAdd(dead, &redis.Z{Member: string(bytes), Score: float64(now.Unix())})
-	limit := now.AddDate(0, 0, -deadExpirationInDays).Unix() // 90 days ago
-	pipe.ZRemRangeByScore(dead, "-inf", strconv.Itoa(int(limit)))
-	pipe.ZRemRangeByRank(dead, 0, -maxDeadTask) // trim the set to 100
-	_, err = pipe.Exec()
-	return err
-}
-
-// listQueues returns the list of all queues.
-func listQueues(rdb *redis.Client) []string {
-	return rdb.SMembers(allQueues).Val()
-}
-
-// delaySeconds returns a number seconds to delay before retrying.
-// Formula taken from https://github.com/mperham/sidekiq.
-func delaySeconds(count int) time.Duration {
-	r := rand.New(rand.NewSource(time.Now().UnixNano()))
-	s := int(math.Pow(float64(count), 4)) + 15 + (r.Intn(30) * (count + 1))
-	return time.Duration(s) * time.Second
-}
diff --git a/manager.go b/manager.go
index 8339549..dad8bee 100644
--- a/manager.go
+++ b/manager.go
@@ -1,16 +1,15 @@
 package asynq
 
 import (
-	"encoding/json"
 	"fmt"
 	"log"
+	"math"
+	"math/rand"
 	"time"
-
-	"github.com/go-redis/redis/v7"
 )
 
 type manager struct {
-	rdb *redis.Client
+	rdb *rdb
 
 	handler TaskHandler
 
@@ -22,7 +21,7 @@ type manager struct {
 	done chan struct{}
 }
 
-func newManager(rdb *redis.Client, numWorkers int, handler TaskHandler) *manager {
+func newManager(rdb *rdb, numWorkers int, handler TaskHandler) *manager {
 	return &manager{
 		rdb:     rdb,
 		handler: handler,
@@ -63,22 +62,21 @@ func (m *manager) start() {
 func (m *manager) processTasks() {
 	// pull message out of the queue and process it
 	// TODO(hibiken): sort the list of queues in order of priority
-	res, err := m.rdb.BLPop(5*time.Second, listQueues(m.rdb)...).Result() // NOTE: BLPOP needs to time out because if case a new queue is added.
+	msg, err := m.rdb.bpop(5*time.Second, m.rdb.listQueues()...)
 	if err != nil {
-		if err != redis.Nil {
-			log.Printf("BLPOP command failed: %v\n", err)
+		switch err {
+		case errQueuePopTimeout:
+			// timed out, this is a normal behavior.
+			return
+		case errDeserializeTask:
+			log.Println("[Servere Error] could not parse json encoded message")
+			return
+		default:
+			log.Printf("[Servere Error] unexpected error while pulling message out of queues: %v\n", err)
+			return
 		}
-		return
 	}
 
-	q, data := res[0], res[1]
-	fmt.Printf("perform task %v from %s\n", data, q)
-	var msg taskMessage
-	err = json.Unmarshal([]byte(data), &msg)
-	if err != nil {
-		log.Printf("[Servere Error] could not parse json encoded message %s: %v", data, err)
-		return
-	}
 	t := &Task{Type: msg.Type, Payload: msg.Payload}
 	m.sema <- struct{}{} // acquire token
 	go func(task *Task) {
@@ -87,7 +85,7 @@ func (m *manager) processTasks() {
 		if err != nil {
 			if msg.Retried >= msg.Retry {
 				fmt.Println("Retry exhausted!!!")
-				if err := kill(m.rdb, &msg); err != nil {
+				if err := m.rdb.kill(msg); err != nil {
 					log.Printf("[SERVER ERROR] could not add task %+v to 'dead' set\n", err)
 				}
 				return
@@ -97,7 +95,7 @@ func (m *manager) processTasks() {
 			fmt.Printf("[DEBUG] retying the task in %v\n", retryAt.Sub(time.Now()))
 			msg.Retried++
 			msg.ErrorMsg = err.Error()
-			if err := zadd(m.rdb, retry, float64(retryAt.Unix()), &msg); err != nil {
+			if err := m.rdb.zadd(retry, float64(retryAt.Unix()), msg); err != nil {
 				// TODO(hibiken): Not sure how to handle this error
 				log.Printf("[SEVERE ERROR] could not add msg %+v to 'retry' set: %v\n", msg, err)
 				return
@@ -105,3 +103,11 @@ func (m *manager) processTasks() {
 		}
 	}(t)
 }
+
+// delaySeconds returns a number seconds to delay before retrying.
+// Formula taken from https://github.com/mperham/sidekiq.
+func delaySeconds(count int) time.Duration {
+	r := rand.New(rand.NewSource(time.Now().UnixNano()))
+	s := int(math.Pow(float64(count), 4)) + 15 + (r.Intn(30) * (count + 1))
+	return time.Duration(s) * time.Second
+}
diff --git a/poller.go b/poller.go
index 5f969ee..7f5452f 100644
--- a/poller.go
+++ b/poller.go
@@ -1,7 +1,6 @@
 package asynq
 
 import (
-	"encoding/json"
 	"fmt"
 	"log"
 	"strconv"
@@ -11,7 +10,7 @@ import (
 )
 
 type poller struct {
-	rdb *redis.Client
+	rdb *rdb
 
 	// channel to communicate back to the long running "poller" goroutine.
 	done chan struct{}
@@ -23,6 +22,15 @@ type poller struct {
 	zsets []string
 }
 
+func newPoller(rdb *rdb, avgInterval time.Duration, zsets []string) *poller {
+	return &poller{
+		rdb:         rdb,
+		done:        make(chan struct{}),
+		avgInterval: avgInterval,
+		zsets:       zsets,
+	}
+}
+
 func (p *poller) terminate() {
 	// send a signal to the manager goroutine to stop
 	// processing tasks from the queue.
@@ -51,38 +59,19 @@ func (p *poller) enqueue() {
 		// Get next items in the queue with scores (time to execute) <= now.
 		now := time.Now().Unix()
 		fmt.Printf("[DEBUG] polling ZSET %q\n", zset)
-		jobs, err := p.rdb.ZRangeByScore(zset,
-			&redis.ZRangeBy{
-				Min: "-inf",
-				Max: strconv.Itoa(int(now))}).Result()
-		fmt.Printf("len(jobs) = %d\n", len(jobs))
+		msgs, err := p.rdb.zRangeByScore(zset,
+			&redis.ZRangeBy{Min: "-inf", Max: strconv.Itoa(int(now))})
 		if err != nil {
 			log.Printf("radis command ZRANGEBYSCORE failed: %v\n", err)
 			continue
 		}
-		if len(jobs) == 0 {
-			fmt.Println("jobs empty")
-			continue
-		}
 
-		for _, j := range jobs {
-			fmt.Printf("[debug] j = %v\n", j)
-			var msg taskMessage
-			err = json.Unmarshal([]byte(j), &msg)
-			if err != nil {
-				fmt.Println("unmarshal failed")
+		for _, m := range msgs {
+			if err := p.rdb.move(zset, m); err != nil {
+				log.Printf("could not move task %+v to queue %q: %v",
+					m, m.Queue, err)
 				continue
 			}
-
-			fmt.Println("[debug] ZREM")
-			if p.rdb.ZRem(zset, j).Val() > 0 {
-				err = push(p.rdb, &msg)
-				if err != nil {
-					log.Printf("could not push task to queue %q: %v", msg.Queue, err)
-					// TODO(hibiken): Handle this error properly. Add back to scheduled ZSET?
-					continue
-				}
-			}
 		}
 	}
 }
diff --git a/rdb.go b/rdb.go
new file mode 100644
index 0000000..d2371bf
--- /dev/null
+++ b/rdb.go
@@ -0,0 +1,155 @@
+package asynq
+
+import (
+	"encoding/json"
+	"errors"
+	"fmt"
+	"log"
+	"strconv"
+	"time"
+
+	"github.com/go-redis/redis/v7"
+)
+
+// Redis keys
+const (
+	queuePrefix = "asynq:queues:"   // LIST
+	allQueues   = "asynq:queues"    // SET
+	scheduled   = "asynq:scheduled" // ZSET
+	retry       = "asynq:retry"     // ZSET
+	dead        = "asynq:dead"      // ZSET
+)
+
+var (
+	errQueuePopTimeout = errors.New("blocking queue pop operation timed out")
+	errSerializeTask   = errors.New("could not encode task message into json")
+	errDeserializeTask = errors.New("could not decode task message from json")
+)
+
+// rdb encapsulates the interaction with redis server.
+type rdb struct {
+	client *redis.Client
+}
+
+func newRDB(client *redis.Client) *rdb {
+	return &rdb{client}
+}
+
+// push enqueues the task to queue.
+func (r *rdb) push(msg *taskMessage) error {
+	bytes, err := json.Marshal(msg)
+	if err != nil {
+		return fmt.Errorf("could not encode task into JSON: %v", err)
+	}
+	qname := queuePrefix + msg.Queue
+	err = r.client.SAdd(allQueues, qname).Err()
+	if err != nil {
+		return fmt.Errorf("command SADD %q %q failed: %v",
+			allQueues, qname, err)
+	}
+	err = r.client.RPush(qname, string(bytes)).Err()
+	if err != nil {
+		return fmt.Errorf("command RPUSH %q %q failed: %v",
+			qname, string(bytes), err)
+	}
+	return nil
+}
+
+// bpop blocks until there is a taskMessage available to be processed.
+// bpop returns immediately if there are already taskMessages waiting to be processed.
+func (r *rdb) bpop(timeout time.Duration, keys ...string) (*taskMessage, error) {
+	res, err := r.client.BLPop(5*time.Second, keys...).Result() // NOTE: BLPOP needs to time out because if case a new queue is added.
+	if err != nil {
+		if err != redis.Nil {
+			return nil, fmt.Errorf("command BLPOP %v %v failed: %v",
+				timeout, keys, err)
+		}
+		return nil, errQueuePopTimeout
+	}
+	q, data := res[0], res[1]
+	fmt.Printf("perform task %v from %s\n", data, q)
+	var msg taskMessage
+	err = json.Unmarshal([]byte(data), &msg)
+	if err != nil {
+		return nil, errDeserializeTask
+	}
+	return &msg, nil
+}
+
+// zadd adds the taskMessage to the specified zset (sorted set) with the given score.
+func (r *rdb) zadd(zset string, zscore float64, msg *taskMessage) error {
+	bytes, err := json.Marshal(msg)
+	if err != nil {
+		return fmt.Errorf("could not encode task into JSON: %v", err)
+	}
+	err = r.client.ZAdd(zset, &redis.Z{Member: string(bytes), Score: zscore}).Err()
+	if err != nil {
+		return fmt.Errorf("command ZADD %s %.1f %s failed: %v",
+			zset, zscore, string(bytes), err)
+	}
+	return nil
+}
+
+func (r *rdb) zRangeByScore(key string, opt *redis.ZRangeBy) ([]*taskMessage, error) {
+	jobs, err := r.client.ZRangeByScore(key, opt).Result()
+	fmt.Printf("len(jobs) = %d\n", len(jobs))
+	if err != nil {
+		return nil, fmt.Errorf("command ZRANGEBYSCORE %s %v failed: %v", key, opt, err)
+	}
+	var msgs []*taskMessage
+	for _, j := range jobs {
+		fmt.Printf("[debug] j = %v\n", j)
+		var msg taskMessage
+		err = json.Unmarshal([]byte(j), &msg)
+		if err != nil {
+			log.Printf("[WARNING] could not unmarshal task data %s: %v\n", j, err)
+			continue
+		}
+		msgs = append(msgs, &msg)
+	}
+	return msgs, nil
+}
+
+// move moves taskMessage from zfrom to the specified queue.
+func (r *rdb) move(from string, msg *taskMessage) error {
+	bytes, err := json.Marshal(msg)
+	if err != nil {
+		return errSerializeTask
+	}
+	if r.client.ZRem(from, string(bytes)).Val() > 0 {
+		err = r.push(msg)
+		if err != nil {
+			log.Printf("[SERVERE ERROR] could not push task to queue %q: %v\n",
+				msg.Queue, err)
+			// TODO(hibiken): Handle this error properly.
+			// Add back to zfrom?
+			return fmt.Errorf("could not push task %v from %q: %v", msg, msg.Queue, err)
+		}
+	}
+	return nil
+}
+
+const maxDeadTask = 100
+const deadExpirationInDays = 90
+
+// kill sends the taskMessage to "dead" set.
+// It also trims the sorted set by timestamp and set size.
+func (r *rdb) kill(msg *taskMessage) error {
+	bytes, err := json.Marshal(msg)
+	if err != nil {
+		return fmt.Errorf("could not encode task into JSON: %v", err)
+	}
+	now := time.Now()
+	pipe := r.client.Pipeline()
+	pipe.ZAdd(dead, &redis.Z{Member: string(bytes), Score: float64(now.Unix())})
+	limit := now.AddDate(0, 0, -deadExpirationInDays).Unix() // 90 days ago
+	pipe.ZRemRangeByScore(dead, "-inf", strconv.Itoa(int(limit)))
+	pipe.ZRemRangeByRank(dead, 0, -maxDeadTask) // trim the set to 100
+	_, err = pipe.Exec()
+	return err
+}
+
+// listQueues returns the list of all queues.
+func (r *rdb) listQueues() []string {
+	return r.client.SMembers(allQueues).Val()
+}