diff --git a/cli/cmd/get.go b/cli/cmd/get.go
index 40a61fe133..8905d90006 100644
--- a/cli/cmd/get.go
+++ b/cli/cmd/get.go
@@ -542,6 +542,10 @@ func dataResourceTable(resources []context.Resource, dataStatuses map[string]*re
 func apiResourceTable(apiGroupStatuses map[string]*resource.APIGroupStatus) string {
 	rows := make([][]interface{}, 0, len(apiGroupStatuses))
 	for name, groupStatus := range apiGroupStatuses {
+		if groupStatus.Requested == 0 {
+			continue
+		}
+
 		var updatedAt *time.Time
 		if groupStatus.ActiveStatus != nil {
 			updatedAt = groupStatus.ActiveStatus.Start
diff --git a/pkg/operator/workloads/logs.go b/pkg/operator/workloads/logs.go
index 10a4333540..f0e7b9a51a 100644
--- a/pkg/operator/workloads/logs.go
+++ b/pkg/operator/workloads/logs.go
@@ -18,6 +18,8 @@ package workloads
 
 import (
 	"bufio"
+	"bytes"
+	"encoding/json"
 	"fmt"
 	"io"
 	"os"
@@ -26,19 +28,22 @@ import (
 	"time"
 
 	"github.com/gorilla/websocket"
-
 	kcore "k8s.io/api/core/v1"
 
 	"github.com/cortexlabs/cortex/pkg/lib/errors"
 	"github.com/cortexlabs/cortex/pkg/lib/k8s"
 	"github.com/cortexlabs/cortex/pkg/lib/pointer"
+	"github.com/cortexlabs/cortex/pkg/lib/sets/strset"
 	"github.com/cortexlabs/cortex/pkg/operator/config"
 )
 
 const (
-	writeWait        = 10 * time.Second
-	closeGracePeriod = 10 * time.Second
-	maxMessageSize   = 8192
+	writeWait             = 10 * time.Second
+	closeGracePeriod      = 10 * time.Second
+	maxMessageSize        = 8192
+	podCheckInterval      = 5 * time.Second
+	maxParallelPodLogging = 5
+	initLogTailLines      = 20
 )
 
 func ReadLogs(appName string, workloadID string, verbose bool, socket *websocket.Conn) {
@@ -56,40 +61,46 @@ func ReadLogs(appName string, workloadID string, verbose bool, socket *websocket
 		}
 
 		if len(pods) > 0 {
-			if len(pods) > 1 {
-				if !writeSocket(fmt.Sprintf("%d pods available, streaming logs for one of them:", len(pods)), socket) {
+			if len(pods) > maxParallelPodLogging {
+				if !writeSocket(fmt.Sprintf("\n%d pods available, streaming logs for %d of them:", len(pods), maxParallelPodLogging), socket) {
 					return
 				}
 			}
 
 			podMap := make(map[k8s.PodStatus][]kcore.Pod)
 			for _, pod := range pods {
-				podMap[k8s.GetPodStatus(&pod)] = append(podMap[k8s.GetPodStatus(&pod)], pod)
+				podStatus := k8s.GetPodStatus(&pod)
+				if len(podMap[podStatus]) < maxParallelPodLogging {
+					podMap[podStatus] = append(podMap[podStatus], pod)
+				}
 			}
 
 			switch {
 			case len(podMap[k8s.PodStatusSucceeded]) > 0:
-				getKubectlLogs(&podMap[k8s.PodStatusSucceeded][0], verbose, wrotePending, false, socket)
+				getKubectlLogs(podMap[k8s.PodStatusSucceeded], verbose, wrotePending, false, socket)
 			case len(podMap[k8s.PodStatusRunning]) > 0:
-				getKubectlLogs(&podMap[k8s.PodStatusRunning][0], verbose, wrotePending, false, socket)
+				getKubectlLogs(podMap[k8s.PodStatusRunning], verbose, wrotePending, false, socket)
 			case len(podMap[k8s.PodStatusPending]) > 0:
-				getKubectlLogs(&podMap[k8s.PodStatusPending][0], verbose, wrotePending, false, socket)
+				getKubectlLogs(podMap[k8s.PodStatusPending], verbose, wrotePending, false, socket)
 			case len(podMap[k8s.PodStatusKilled]) > 0:
-				getKubectlLogs(&podMap[k8s.PodStatusKilled][0], verbose, wrotePending, false, socket)
+				getKubectlLogs(podMap[k8s.PodStatusKilled], verbose, wrotePending, false, socket)
 			case len(podMap[k8s.PodStatusKilledOOM]) > 0:
-				getKubectlLogs(&podMap[k8s.PodStatusKilledOOM][0], verbose, wrotePending, false, socket)
+				getKubectlLogs(podMap[k8s.PodStatusKilledOOM], verbose, wrotePending, false, socket)
 			case len(podMap[k8s.PodStatusFailed]) > 0:
 				previous := false
 				if pods[0].Labels["workloadType"] == WorkloadTypeAPI {
 					previous = true
 				}
-				getKubectlLogs(&podMap[k8s.PodStatusFailed][0], verbose, wrotePending, previous, socket)
+				getKubectlLogs(podMap[k8s.PodStatusFailed], verbose, wrotePending, previous, socket)
 			case len(podMap[k8s.PodStatusTerminating]) > 0:
-				getKubectlLogs(&podMap[k8s.PodStatusTerminating][0], verbose, wrotePending, false, socket)
+				getKubectlLogs(podMap[k8s.PodStatusTerminating], verbose, wrotePending, false, socket)
 			case len(podMap[k8s.PodStatusUnknown]) > 0:
-				getKubectlLogs(&podMap[k8s.PodStatusUnknown][0], verbose, wrotePending, false, socket)
+				getKubectlLogs(podMap[k8s.PodStatusUnknown], verbose, wrotePending, false, socket)
 			default: // unexpected
-				getKubectlLogs(&pods[0], verbose, wrotePending, false, socket)
+				if len(pods) > maxParallelPodLogging {
+					pods = pods[:maxParallelPodLogging]
+				}
+				getKubectlLogs(pods, verbose, wrotePending, false, socket)
 			}
 			return
 		}
@@ -118,7 +129,7 @@ func ReadLogs(appName string, workloadID string, verbose bool, socket *websocket
 			if !writeSocket("\nFailed to start:\n", socket) {
 				return
 			}
-			getKubectlLogs(failedArgoPod, true, false, false, socket)
+			getKubectlLogs([]kcore.Pod{*failedArgoPod}, true, false, false, socket)
 			return
 		}
 
@@ -133,52 +144,169 @@ func ReadLogs(appName string, workloadID string, verbose bool, socket *websocket
 	}
 }
 
-func getKubectlLogs(pod *kcore.Pod, verbose bool, wrotePending bool, previous bool, socket *websocket.Conn) {
-	cmdPath := "/usr/local/bin/kubectl"
+func getKubectlLogs(pods []kcore.Pod, verbose bool, wrotePending bool, previous bool, socket *websocket.Conn) {
+	isAllPending := true
+	for _, pod := range pods {
+		if k8s.GetPodStatus(&pod) != k8s.PodStatusPending {
+			isAllPending = false
+			break
+		}
+	}
 
-	if k8s.GetPodStatus(pod) == k8s.PodStatusPending {
-		if !wrotePending {
-			if !writeSocket("\nPending", socket) {
-				return
-			}
+	if isAllPending {
+		if !writeSocket("\nPending", socket) {
+			return
 		}
-		config.Kubernetes.WaitForPodRunning(pod.Name, 1)
 	}
 
-	args := []string{"kubectl", "-n=" + config.Cortex.Namespace, "logs", "--follow=true"}
+	inr, inw, err := os.Pipe()
+	if err != nil {
+		errors.Panic(err, "logs", "kubectl", "os.pipe")
+	}
+	defer inw.Close()
+	defer inr.Close()
+
+	podCheckCancel := make(chan struct{})
+	defer close(podCheckCancel)
+
+	go podCheck(podCheckCancel, socket, pods, previous, verbose, inr)
+	pumpStdin(socket, inw)
+	podCheckCancel <- struct{}{}
+}
+
+func startKubectlProcess(pod kcore.Pod, previous bool, attrs *os.ProcAttr) (*os.Process, error) {
+	cmdPath := "/bin/bash"
+
+	kubectlArgs := []string{"kubectl", "-n=" + config.Cortex.Namespace, "logs", "--follow=true"}
 	if previous {
-		args = append(args, "--previous")
+		kubectlArgs = append(kubectlArgs, "--previous")
 	}
 
-	args = append(args, pod.Name)
+	identifier := pod.Name
+	kubectlArgs = append(kubectlArgs, pod.Name)
 	if pod.Labels["workloadType"] == WorkloadTypeAPI && pod.Labels["userFacing"] == "true" {
-		args = append(args, apiContainerName)
+		kubectlArgs = append(kubectlArgs, apiContainerName)
+		kubectlArgs = append(kubectlArgs, fmt.Sprintf("--tail=%d", initLogTailLines))
+		identifier += " " + apiContainerName
 	}
 
-	outr, outw, err := os.Pipe()
+	labelLog := fmt.Sprintf(" | while read -r; do echo \"[%s] $REPLY \" | tail -n +1; done", identifier)
+	kubectlCmd := strings.Join(kubectlArgs, " ")
+	bashArgs := []string{"/bin/bash", "-c", kubectlCmd + labelLog}
+	process, err := os.StartProcess(cmdPath, bashArgs, attrs)
 	if err != nil {
-		errors.Panic(err, "logs", "kubectl", "os.pipe")
+		return nil, errors.Wrap(err, strings.Join(bashArgs, " "))
 	}
-	defer outr.Close()
-	defer outw.Close()
 
-	inr, inw, err := os.Pipe()
+	return process, nil
+}
+
+func podCheck(podCheckCancel chan struct{}, socket *websocket.Conn, initialPodList []kcore.Pod, previous bool, verbose bool, inr *os.File) {
+	timer := time.NewTimer(0)
+	defer timer.Stop()
+
+	processMap := make(map[string]*os.Process)
+	defer deleteProcesses(processMap)
+	labels := initialPodList[0].GetLabels()
+	appName := labels["appName"]
+	workloadID := labels["workloadID"]
+
+	outr, outw, err := os.Pipe()
 	if err != nil {
 		errors.Panic(err, "logs", "kubectl", "os.pipe")
 	}
-	defer inr.Close()
-	defer inw.Close()
+	defer outw.Close()
+	defer outr.Close()
 
-	process, err := os.StartProcess(cmdPath, args, &os.ProcAttr{
-		Files: []*os.File{inr, outw, outw},
-	})
-	if err != nil {
-		errors.Panic(err, strings.Join(args, " "))
+	socketWriterError := make(chan error, 1)
+	defer close(socketWriterError)
+
+	go pumpStdout(socket, socketWriterError, outr, verbose, true)
+
+	for {
+		select {
+		case <-podCheckCancel:
+			return
+		case <-timer.C:
+			pods, err := config.Kubernetes.ListPodsByLabels(map[string]string{
+				"appName":    appName,
+				"workloadID": workloadID,
+				"userFacing": "true",
+			})
+
+			if err != nil {
+				socketWriterError <- errors.Wrap(err, "pod check")
+				return
+			}
+
+			latestRunningPodsMap := make(map[string]kcore.Pod)
+			latestRunningPods := strset.New()
+			for _, pod := range pods {
+				if k8s.GetPodStatus(&pod) != k8s.PodStatusPending {
+					latestRunningPods.Add(pod.GetName())
+					latestRunningPodsMap[pod.GetName()] = pod
+				}
+			}
+
+			prevRunningPods := strset.New()
+			for podName := range processMap {
+				prevRunningPods.Add(podName)
+			}
+
+			newPods := strset.Difference(latestRunningPods, prevRunningPods)
+			podsToDelete := strset.Difference(prevRunningPods, latestRunningPods)
+			podsToKeep := strset.Intersection(prevRunningPods, latestRunningPods)
+
+			// Prioritize adding running pods
+			podsToAddRunning := []string{}
+			podsToAddNotRunning := []string{}
+
+			for podName := range newPods {
+				pod := latestRunningPodsMap[podName]
+				if k8s.GetPodStatus(&pod) == k8s.PodStatusRunning {
+					podsToAddRunning = append(podsToAddRunning, podName)
+				} else {
+					podsToAddNotRunning = append(podsToAddNotRunning, podName)
+				}
+			}
+			podsToAdd := append(podsToAddRunning, podsToAddNotRunning...)
+
+			maxPodsToAdd := maxParallelPodLogging - len(podsToKeep)
+			if len(podsToAdd) < maxPodsToAdd {
+				maxPodsToAdd = len(podsToAdd)
+			}
+
+			for _, podName := range podsToAdd[:maxPodsToAdd] {
+				process, err := startKubectlProcess(latestRunningPodsMap[podName], previous, &os.ProcAttr{
+					Files: []*os.File{inr, outw, outw},
+				})
+				if err != nil {
+					socketWriterError <- err
+					return
+				}
+				processMap[podName] = process
+			}
+
+			deleteMap := make(map[string]*os.Process, len(podsToDelete))
+
+			for podName := range podsToDelete {
+				deleteMap[podName] = processMap[podName]
+				delete(processMap, podName)
+			}
+			deleteProcesses(deleteMap)
+			timer.Reset(podCheckInterval)
+		}
 	}
+}
 
-	go pumpStdout(socket, outr, verbose, true)
-	pumpStdin(socket, inw)
-	stopProcess(process)
+func deleteProcesses(processMap map[string]*os.Process) {
+	for _, process := range processMap {
+		process.Signal(os.Interrupt)
+	}
+	time.Sleep(5 * time.Second)
+	for _, process := range processMap {
+		process.Signal(os.Kill)
+	}
 }
 
 func getCloudWatchLogs(prefix string, verbose bool, socket *websocket.Conn) {
@@ -196,7 +324,10 @@ func getCloudWatchLogs(prefix string, verbose bool, socket *websocket.Conn) {
 	} else {
 		logsReader = strings.NewReader(logs)
 	}
-	go pumpStdout(socket, logsReader, verbose, false)
+
+	socketWriterError := make(chan error)
+	defer close(socketWriterError)
+	go pumpStdout(socket, socketWriterError, logsReader, verbose, false)
 
 	inr, inw, err := os.Pipe()
 	if err != nil {
@@ -224,7 +355,7 @@ func pumpStdin(socket *websocket.Conn, writer io.Writer) {
 	}
 }
 
-func pumpStdout(socket *websocket.Conn, reader io.Reader, verbose bool, checkForLastLog bool) {
+func pumpStdout(socket *websocket.Conn, socketWriterError chan error, reader io.Reader, verbose bool, checkForLastLog bool) {
 	scanner := bufio.NewScanner(reader)
 	for scanner.Scan() {
 		socket.SetWriteDeadline(time.Now().Add(writeWait))
@@ -243,14 +374,23 @@ func pumpStdout(socket *websocket.Conn, reader io.Reader, verbose bool, checkFor
 		}
 	}
 
+	select {
+	case err := <-socketWriterError:
+		if err != nil {
+			writeSocket(err.Error(), socket)
+		}
+	default:
+	}
+
 	socket.SetWriteDeadline(time.Now().Add(writeWait))
 	socket.WriteMessage(websocket.CloseMessage, websocket.FormatCloseMessage(websocket.CloseNormalClosure, ""))
 	time.Sleep(closeGracePeriod)
 	socket.Close()
 }
 
-var cortexRegex = regexp.MustCompile(`^?(DEBUG|INFO|WARNING|ERROR|CRITICAL):cortex:`)
-var tensorflowRegex = regexp.MustCompile(`^?(DEBUG|INFO|WARNING|ERROR|CRITICAL):tensorflow:`)
+var cortexRegex = regexp.MustCompile(`^\[.*\]?(DEBUG|INFO|WARNING|ERROR|CRITICAL):cortex:`)
+var tensorflowRegex = regexp.MustCompile(`^\[.*\]?(DEBUG|INFO|WARNING|ERROR|CRITICAL):tensorflow:`)
+var jsonPrefixRegex = regexp.MustCompile(`^\ *?(\{|\[)`)
 
 func formatHeader1(headerString string) *string {
 	headerBorder := "\n" + strings.Repeat("-", len(headerString)) + "\n"
@@ -299,6 +439,20 @@ func extractFromCortexLog(match string, loglevel string, logStr string) (*string
 		return formatHeader3(cutStr), false
 	}
 
+	matches := jsonPrefixRegex.FindStringSubmatch(cutStr)
+	if len(matches) == 2 {
+		indentIndex := len(matches[0]) - 1 // matches to curly or square bracket so remove the last char
+		indentStr := cutStr[:indentIndex]
+		maybeJSON := cutStr[indentIndex:]
+		jsonBytes := []byte(maybeJSON)
+		var obytes bytes.Buffer
+		err := json.Indent(&obytes, jsonBytes, indentStr, "  ")
+		if err == nil {
+			ostr := indentStr + string(obytes.String())
+			return &ostr, false
+		}
+	}
+
 	lastLogRe := regexp.MustCompile(`^workload: (\w+), completed: (\S+)`)
 	if lastLogRe.MatchString(cutStr) {
 		return &cutStr, true
diff --git a/pkg/workloads/cortex/lib/util.py b/pkg/workloads/cortex/lib/util.py
index 09a16d0c0e..808fafdf87 100644
--- a/pkg/workloads/cortex/lib/util.py
+++ b/pkg/workloads/cortex/lib/util.py
@@ -118,7 +118,7 @@ def log_pretty(obj, indent=0, logging_func=logger.info):
 
 
 def log_pretty_flat(obj, indent=0, logging_func=logger.info):
-    logging_func(pp_str_flat(obj), indent)
+    logging_func(pp_str_flat(obj, indent))
 
 
 def pluralize(num, singular, plural):
diff --git a/pkg/workloads/cortex/onnx_serve/api.py b/pkg/workloads/cortex/onnx_serve/api.py
index de2267a153..e02d99f673 100644
--- a/pkg/workloads/cortex/onnx_serve/api.py
+++ b/pkg/workloads/cortex/onnx_serve/api.py
@@ -135,7 +135,7 @@ def predict(app_name, api_name):
     response = {}
 
     if not util.is_dict(payload) or "samples" not in payload:
-        util.log_pretty(payload, logging_func=logger.error)
+        util.log_pretty_flat(payload, logging_func=logger.error)
         return prediction_failed(payload, "top level `samples` key not found in request")
 
     logger.info("Predicting " + util.pluralize(len(payload["samples"]), "sample", "samples"))
@@ -143,7 +143,7 @@ def predict(app_name, api_name):
     predictions = []
     samples = payload["samples"]
     if not util.is_list(samples):
-        util.log_pretty(samples, logging_func=logger.error)
+        util.log_pretty_flat(samples, logging_func=logger.error)
         return prediction_failed(
             payload, "expected the value of key `samples` to be a list of json objects"
         )
@@ -152,7 +152,7 @@ def predict(app_name, api_name):
         util.log_indent("sample {}".format(i + 1), 2)
         try:
             util.log_indent("Raw sample:", indent=4)
-            util.log_pretty(sample, indent=6)
+            util.log_pretty_flat(sample, indent=6)
 
             if request_handler is not None and util.has_function(request_handler, "pre_inference"):
                 sample = request_handler.pre_inference(sample, input_metadata)
@@ -169,7 +169,7 @@ def predict(app_name, api_name):
             if request_handler is not None and util.has_function(request_handler, "post_inference"):
                 result = request_handler.post_inference(result, output_metadata)
             util.log_indent("Prediction:", indent=4)
-            util.log_pretty(result, indent=6)
+            util.log_pretty_flat(result, indent=6)
             prediction = {"prediction": result}
         except CortexException as e:
             e.wrap("error", "sample {}".format(i + 1))
diff --git a/pkg/workloads/cortex/tf_api/api.py b/pkg/workloads/cortex/tf_api/api.py
index f6a56eeea2..4f40c5062f 100644
--- a/pkg/workloads/cortex/tf_api/api.py
+++ b/pkg/workloads/cortex/tf_api/api.py
@@ -259,11 +259,11 @@ def run_predict(sample):
         result = parse_response_proto(response_proto)
 
         util.log_indent("Raw sample:", indent=4)
-        util.log_pretty(sample, indent=6)
+        util.log_pretty_flat(sample, indent=6)
         util.log_indent("Transformed sample:", indent=4)
-        util.log_pretty(transformed_sample, indent=6)
+        util.log_pretty_flat(transformed_sample, indent=6)
         util.log_indent("Prediction:", indent=4)
-        util.log_pretty(result, indent=6)
+        util.log_pretty_flat(result, indent=6)
 
         result["transformed_sample"] = transformed_sample
 
@@ -272,9 +272,9 @@ def run_predict(sample):
         response_proto = local_cache["stub"].Predict(prediction_request, timeout=10.0)
         result = parse_response_proto_raw(response_proto)
         util.log_indent("Sample:", indent=4)
-        util.log_pretty(sample, indent=6)
+        util.log_pretty_flat(sample, indent=6)
         util.log_indent("Prediction:", indent=4)
-        util.log_pretty(result, indent=6)
+        util.log_pretty_flat(result, indent=6)
 
     if request_handler is not None and util.has_function(request_handler, "post_inference"):
         result = request_handler.post_inference(result, local_cache["metadata"]["signatureDef"])
@@ -315,6 +315,7 @@ def health():
 
 @app.route("/<deployment_name>/<api_name>", methods=["POST"])
 def predict(deployment_name, api_name):
+
     try:
         payload = request.get_json()
     except Exception as e:
@@ -326,7 +327,7 @@ def predict(deployment_name, api_name):
     response = {}
 
     if not util.is_dict(payload) or "samples" not in payload:
-        util.log_pretty(payload, logging_func=logger.error)
+        util.log_pretty_flat(payload, logging_func=logger.error)
         return prediction_failed(payload, "top level `samples` key not found in request")
 
     logger.info("Predicting " + util.pluralize(len(payload["samples"]), "sample", "samples"))
@@ -334,7 +335,7 @@ def predict(deployment_name, api_name):
     predictions = []
     samples = payload["samples"]
     if not util.is_list(samples):
-        util.log_pretty(samples, logging_func=logger.error)
+        util.log_pretty_flat(samples, logging_func=logger.error)
         return prediction_failed(
             payload, "expected the value of key `samples` to be a list of json objects"
         )