consul/test/integration/connect/envoy/helpers.bash

641 lines
15 KiB
Bash
Raw Normal View History

#!/bin/bash
# retry based on
# https://github.com/fernandoacorreia/azure-docker-registry/blob/master/tools/scripts/create-registry-server
# under MIT license.
function retry {
local n=1
local max=$1
shift
local delay=$1
shift
local errtrace=0
if grep -q "errtrace" <<< "$SHELLOPTS"
then
errtrace=1
set +E
fi
for ((i=1;i<=$max;i++))
do
if "$@"
then
if test $errtrace -eq 1
then
set -E
fi
return 0
else
echo "Command failed. Attempt $i/$max:"
sleep $delay
fi
done
if test $errtrace -eq 1
then
set -E
fi
return 1
}
function retry_default {
set +E
ret=0
retry 5 1 "$@" || ret=1
set -E
return $ret
}
function retry_long {
retry 30 1 "$@"
}
function echored {
tput setaf 1
tput bold
echo $@
tput sgr0
}
function echogreen {
tput setaf 2
tput bold
echo $@
tput sgr0
}
function echoyellow {
tput setaf 3
tput bold
echo $@
tput sgr0
}
function echoblue {
tput setaf 4
tput bold
echo $@
tput sgr0
}
function is_set {
# Arguments:
# $1 - string value to check its truthiness
#
# Return:
# 0 - is truthy (backwards I know but allows syntax like `if is_set <var>` to work)
# 1 - is not truthy
local val=$(tr '[:upper:]' '[:lower:]' <<< "$1")
case $val in
1 | t | true | y | yes)
return 0
;;
*)
return 1
;;
esac
}
function get_cert {
local HOSTPORT=$1
CERT=$(openssl s_client -connect $HOSTPORT -showcerts )
openssl x509 -noout -text <<< "$CERT"
}
function assert_proxy_presents_cert_uri {
local HOSTPORT=$1
local SERVICENAME=$2
local DC=${3:-primary}
local NS=${4:-default}
CERT=$(retry_default get_cert $HOSTPORT)
echo "WANT SERVICE: ${NS}/${SERVICENAME}"
echo "GOT CERT:"
echo "$CERT"
echo "$CERT" | grep -Eo "URI:spiffe://([a-zA-Z0-9-]+).consul/ns/${NS}/dc/${DC}/svc/$SERVICENAME"
}
function assert_envoy_version {
local ADMINPORT=$1
run retry_default curl -f -s localhost:$ADMINPORT/server_info
[ "$status" -eq 0 ]
# Envoy 1.8.0 returns a plain text line like
# envoy 5d25f466c3410c0dfa735d7d4358beb76b2da507/1.8.0/Clean/DEBUG live 3 3 0
# Later versions return JSON.
if (echo $output | grep '^envoy') ; then
VERSION=$(echo $output | cut -d ' ' -f 2)
else
VERSION=$(echo $output | jq -r '.version')
fi
echo "Status=$status"
echo "Output=$output"
echo "---"
echo "Got version=$VERSION"
echo "Want version=$ENVOY_VERSION"
echo $VERSION | grep "/$ENVOY_VERSION/"
}
function get_envoy_listener_filters {
local HOSTPORT=$1
run retry_default curl -s -f $HOSTPORT/config_dump
[ "$status" -eq 0 ]
local ENVOY_VERSION=$(echo $output | jq --raw-output '.configs[0].bootstrap.node.metadata.envoy_version')
local QUERY=''
if [ "$ENVOY_VERSION" == "1.13.0" ]; then
QUERY='.configs[2].dynamic_listeners[].active_state.listener | "\(.name) \( .filter_chains[0].filters | map(.name) | join(","))"'
else
QUERY='.configs[2].dynamic_active_listeners[].listener | "\(.name) \( .filter_chains[0].filters | map(.name) | join(","))"'
fi
echo "$output" | jq --raw-output "$QUERY"
}
function get_envoy_cluster_threshold {
local HOSTPORT=$1
local CLUSTER_NAME=$2
run retry_default curl -s -f $HOSTPORT/config_dump
[ "$status" -eq 0 ]
echo "$output" | jq --raw-output "
.configs[1].dynamic_active_clusters[]
| select(.cluster.name|startswith(\"${CLUSTER_NAME}\"))
| .cluster.circuit_breakers.thresholds[0]
"
}
function get_envoy_stats_flush_interval {
local HOSTPORT=$1
run retry_default curl -s -f $HOSTPORT/config_dump
[ "$status" -eq 0 ]
#echo "$output" > /workdir/s1_envoy_dump.json
echo "$output" | jq --raw-output '.configs[0].bootstrap.stats_flush_interval'
}
# snapshot_envoy_admin is meant to be used from a teardown scriptlet from the host.
function snapshot_envoy_admin {
local HOSTPORT=$1
local ENVOY_NAME=$2
local DC=${3:-primary}
local OUTDIR="${LOG_DIR}/envoy-snapshots/${DC}/${ENVOY_NAME}"
mkdir -p "${OUTDIR}"
docker_wget "$DC" "http://${HOSTPORT}/config_dump" -q -O - > "${OUTDIR}/config_dump.json"
docker_wget "$DC" "http://${HOSTPORT}/clusters?format=json" -q -O - > "${OUTDIR}/clusters.json"
docker_wget "$DC" "http://${HOSTPORT}/stats" -q -O - > "${OUTDIR}/stats.txt"
}
function reset_envoy_metrics {
local HOSTPORT=$1
curl -s -f -XPOST $HOSTPORT/reset_counters
return $?
}
function get_all_envoy_metrics {
local HOSTPORT=$1
curl -s -f $HOSTPORT/stats
return $?
}
function get_envoy_metrics {
local HOSTPORT=$1
local METRICS=$2
get_all_envoy_metrics $HOSTPORT | grep "$METRICS"
}
function get_upstream_endpoint_in_status_count {
local HOSTPORT=$1
local CLUSTER_NAME=$2
local HEALTH_STATUS=$3
run retry_default curl -s -f "http://${HOSTPORT}/clusters?format=json"
[ "$status" -eq 0 ]
# echo "$output" >&3
echo "$output" | jq --raw-output "
.cluster_statuses[]
| select(.name|startswith(\"${CLUSTER_NAME}\"))
| [.host_statuses[].health_status.eds_health_status]
| [select(.[] == \"${HEALTH_STATUS}\")]
| length"
}
function assert_upstream_has_endpoints_in_status_once {
local HOSTPORT=$1
local CLUSTER_NAME=$2
local HEALTH_STATUS=$3
local EXPECT_COUNT=$4
GOT_COUNT=$(get_upstream_endpoint_in_status_count $HOSTPORT $CLUSTER_NAME $HEALTH_STATUS)
[ "$GOT_COUNT" -eq $EXPECT_COUNT ]
}
function assert_upstream_has_endpoints_in_status {
local HOSTPORT=$1
local CLUSTER_NAME=$2
local HEALTH_STATUS=$3
local EXPECT_COUNT=$4
run retry_long assert_upstream_has_endpoints_in_status_once $HOSTPORT $CLUSTER_NAME $HEALTH_STATUS $EXPECT_COUNT
[ "$status" -eq 0 ]
}
function assert_envoy_metric {
set -eEuo pipefail
local HOSTPORT=$1
local METRIC=$2
local EXPECT_COUNT=$3
METRICS=$(get_envoy_metrics $HOSTPORT "$METRIC")
if [ -z "${METRICS}" ]
then
echo "Metric not found" 1>&2
return 1
fi
GOT_COUNT=$(awk -F: '{print $2}' <<< "$METRICS" | head -n 1 | tr -d ' ')
if [ -z "$GOT_COUNT" ]
then
echo "Couldn't parse metric count" 1>&2
return 1
fi
if [ $EXPECT_COUNT -ne $GOT_COUNT ]
then
echo "$METRIC - expected count: $EXPECT_COUNT, actual count: $GOT_COUNT" 1>&2
return 1
fi
}
function assert_envoy_metric_at_least {
set -eEuo pipefail
local HOSTPORT=$1
local METRIC=$2
local EXPECT_COUNT=$3
METRICS=$(get_envoy_metrics $HOSTPORT "$METRIC")
if [ -z "${METRICS}" ]
then
echo "Metric not found" 1>&2
return 1
fi
GOT_COUNT=$(awk -F: '{print $2}' <<< "$METRICS" | head -n 1 | tr -d ' ')
if [ -z "$GOT_COUNT" ]
then
echo "Couldn't parse metric count" 1>&2
return 1
fi
if [ $EXPECT_COUNT -gt $GOT_COUNT ]
then
echo "$METRIC - expected >= count: $EXPECT_COUNT, actual count: $GOT_COUNT" 1>&2
return 1
fi
}
function assert_envoy_aggregate_metric_at_least {
set -eEuo pipefail
local HOSTPORT=$1
local METRIC=$2
local EXPECT_COUNT=$3
METRICS=$(get_envoy_metrics $HOSTPORT "$METRIC")
if [ -z "${METRICS}" ]
then
echo "Metric not found" 1>&2
return 1
fi
GOT_COUNT=$(awk '{ sum += $2 } END { print sum }' <<< "$METRICS")
if [ -z "$GOT_COUNT" ]
then
echo "Couldn't parse metric count" 1>&2
return 1
fi
if [ $EXPECT_COUNT -gt $GOT_COUNT ]
then
echo "$METRIC - expected >= count: $EXPECT_COUNT, actual count: $GOT_COUNT" 1>&2
return 1
fi
}
test: for envoy integration tests, wait until 's2' is healthy in consul before interrogating envoy (#6108) When the envoy healthy panic threshold was explicitly disabled as part of L7 traffic management it changed how envoy decided to load balance to endpoints in a cluster. This only matters when envoy is in "panic mode" aka "when you have a bunch of unhealthy endpoints". Panic mode sends traffic to unhealthy instances in certain circumstances. Note: Prior to explicitly disabling the healthy panic threshold, the default value is 50%. What was happening is that the test harness was bringing up consul the sidecars, and the service instances all at once and sometimes the proxies wouldn't have time to be checked by consul to be labeled as 'passing' in the catalog before a round of EDS happened. The xDS server in consul effectively queries /v1/health/connect/s2 and gets 1 result, but that one result has a 'critical' check so the xDS server sends back that endpoint labeled as UNHEALTHY. Envoy sees that 100% of the endpoints in the cluster are unhealthy and would enter panic mode and still send traffic to s2. This is why the test suites PRIOR to disabling the healthy panic threshold worked. They were _incorrectly_ passing. When the healthy panic threshol is disabled, envoy never enters panic mode in this situation and thus the cluster has zero healthy endpoints so load balancing goes nowhere and the tests fail. Why does this only affect the test suites for envoy 1.8.0? My guess is that https://github.com/envoyproxy/envoy/pull/4442 was merged into the 1.9.x series and somehow that plays a role. This PR modifies the bats scripts to explicitly wait until the upstream sidecar is healthy as measured by /v1/health/connect/s2?passing BEFORE trying to interrogate envoy which should make the tests less racy.
2019-07-10 20:58:25 +00:00
function get_healthy_service_count {
local SERVICE_NAME=$1
connect: fix failover through a mesh gateway to a remote datacenter (#6259) Failover is pushed entirely down to the data plane by creating envoy clusters and putting each successive destination in a different load assignment priority band. For example this shows that normally requests go to 1.2.3.4:8080 but when that fails they go to 6.7.8.9:8080: - name: foo load_assignment: cluster_name: foo policy: overprovisioning_factor: 100000 endpoints: - priority: 0 lb_endpoints: - endpoint: address: socket_address: address: 1.2.3.4 port_value: 8080 - priority: 1 lb_endpoints: - endpoint: address: socket_address: address: 6.7.8.9 port_value: 8080 Mesh gateways route requests based solely on the SNI header tacked onto the TLS layer. Envoy currently only lets you configure the outbound SNI header at the cluster layer. If you try to failover through a mesh gateway you ideally would configure the SNI value per endpoint, but that's not possible in envoy today. This PR introduces a simpler way around the problem for now: 1. We identify any target of failover that will use mesh gateway mode local or remote and then further isolate any resolver node in the compiled discovery chain that has a failover destination set to one of those targets. 2. For each of these resolvers we will perform a small measurement of comparative healths of the endpoints that come back from the health API for the set of primary target and serial failover targets. We walk the list of targets in order and if any endpoint is healthy we return that target, otherwise we move on to the next target. 3. The CDS and EDS endpoints both perform the measurements in (2) for the affected resolver nodes. 4. For CDS this measurement selects which TLS SNI field to use for the cluster (note the cluster is always going to be named for the primary target) 5. For EDS this measurement selects which set of endpoints will populate the cluster. Priority tiered failover is ignored. One of the big downsides to this approach to failover is that the failover detection and correction is going to be controlled by consul rather than deferring that entirely to the data plane as with the prior version. This also means that we are bound to only failover using official health signals and cannot make use of data plane signals like outlier detection to affect failover. In this specific scenario the lack of data plane signals is ok because the effectiveness is already muted by the fact that the ultimate destination endpoints will have their data plane signals scrambled when they pass through the mesh gateway wrapper anyway so we're not losing much. Another related fix is that we now use the endpoint health from the underlying service, not the health of the gateway (regardless of failover mode).
2019-08-05 18:30:35 +00:00
local DC=$2
local NS=$3
run retry_default curl -s -f ${HEADERS} "127.0.0.1:8500/v1/health/connect/${SERVICE_NAME}?dc=${DC}&passing&ns=${NS}"
test: for envoy integration tests, wait until 's2' is healthy in consul before interrogating envoy (#6108) When the envoy healthy panic threshold was explicitly disabled as part of L7 traffic management it changed how envoy decided to load balance to endpoints in a cluster. This only matters when envoy is in "panic mode" aka "when you have a bunch of unhealthy endpoints". Panic mode sends traffic to unhealthy instances in certain circumstances. Note: Prior to explicitly disabling the healthy panic threshold, the default value is 50%. What was happening is that the test harness was bringing up consul the sidecars, and the service instances all at once and sometimes the proxies wouldn't have time to be checked by consul to be labeled as 'passing' in the catalog before a round of EDS happened. The xDS server in consul effectively queries /v1/health/connect/s2 and gets 1 result, but that one result has a 'critical' check so the xDS server sends back that endpoint labeled as UNHEALTHY. Envoy sees that 100% of the endpoints in the cluster are unhealthy and would enter panic mode and still send traffic to s2. This is why the test suites PRIOR to disabling the healthy panic threshold worked. They were _incorrectly_ passing. When the healthy panic threshol is disabled, envoy never enters panic mode in this situation and thus the cluster has zero healthy endpoints so load balancing goes nowhere and the tests fail. Why does this only affect the test suites for envoy 1.8.0? My guess is that https://github.com/envoyproxy/envoy/pull/4442 was merged into the 1.9.x series and somehow that plays a role. This PR modifies the bats scripts to explicitly wait until the upstream sidecar is healthy as measured by /v1/health/connect/s2?passing BEFORE trying to interrogate envoy which should make the tests less racy.
2019-07-10 20:58:25 +00:00
[ "$status" -eq 0 ]
echo "$output" | jq --raw-output '. | length'
}
function assert_service_has_healthy_instances_once {
test: for envoy integration tests, wait until 's2' is healthy in consul before interrogating envoy (#6108) When the envoy healthy panic threshold was explicitly disabled as part of L7 traffic management it changed how envoy decided to load balance to endpoints in a cluster. This only matters when envoy is in "panic mode" aka "when you have a bunch of unhealthy endpoints". Panic mode sends traffic to unhealthy instances in certain circumstances. Note: Prior to explicitly disabling the healthy panic threshold, the default value is 50%. What was happening is that the test harness was bringing up consul the sidecars, and the service instances all at once and sometimes the proxies wouldn't have time to be checked by consul to be labeled as 'passing' in the catalog before a round of EDS happened. The xDS server in consul effectively queries /v1/health/connect/s2 and gets 1 result, but that one result has a 'critical' check so the xDS server sends back that endpoint labeled as UNHEALTHY. Envoy sees that 100% of the endpoints in the cluster are unhealthy and would enter panic mode and still send traffic to s2. This is why the test suites PRIOR to disabling the healthy panic threshold worked. They were _incorrectly_ passing. When the healthy panic threshol is disabled, envoy never enters panic mode in this situation and thus the cluster has zero healthy endpoints so load balancing goes nowhere and the tests fail. Why does this only affect the test suites for envoy 1.8.0? My guess is that https://github.com/envoyproxy/envoy/pull/4442 was merged into the 1.9.x series and somehow that plays a role. This PR modifies the bats scripts to explicitly wait until the upstream sidecar is healthy as measured by /v1/health/connect/s2?passing BEFORE trying to interrogate envoy which should make the tests less racy.
2019-07-10 20:58:25 +00:00
local SERVICE_NAME=$1
local EXPECT_COUNT=$2
connect: fix failover through a mesh gateway to a remote datacenter (#6259) Failover is pushed entirely down to the data plane by creating envoy clusters and putting each successive destination in a different load assignment priority band. For example this shows that normally requests go to 1.2.3.4:8080 but when that fails they go to 6.7.8.9:8080: - name: foo load_assignment: cluster_name: foo policy: overprovisioning_factor: 100000 endpoints: - priority: 0 lb_endpoints: - endpoint: address: socket_address: address: 1.2.3.4 port_value: 8080 - priority: 1 lb_endpoints: - endpoint: address: socket_address: address: 6.7.8.9 port_value: 8080 Mesh gateways route requests based solely on the SNI header tacked onto the TLS layer. Envoy currently only lets you configure the outbound SNI header at the cluster layer. If you try to failover through a mesh gateway you ideally would configure the SNI value per endpoint, but that's not possible in envoy today. This PR introduces a simpler way around the problem for now: 1. We identify any target of failover that will use mesh gateway mode local or remote and then further isolate any resolver node in the compiled discovery chain that has a failover destination set to one of those targets. 2. For each of these resolvers we will perform a small measurement of comparative healths of the endpoints that come back from the health API for the set of primary target and serial failover targets. We walk the list of targets in order and if any endpoint is healthy we return that target, otherwise we move on to the next target. 3. The CDS and EDS endpoints both perform the measurements in (2) for the affected resolver nodes. 4. For CDS this measurement selects which TLS SNI field to use for the cluster (note the cluster is always going to be named for the primary target) 5. For EDS this measurement selects which set of endpoints will populate the cluster. Priority tiered failover is ignored. One of the big downsides to this approach to failover is that the failover detection and correction is going to be controlled by consul rather than deferring that entirely to the data plane as with the prior version. This also means that we are bound to only failover using official health signals and cannot make use of data plane signals like outlier detection to affect failover. In this specific scenario the lack of data plane signals is ok because the effectiveness is already muted by the fact that the ultimate destination endpoints will have their data plane signals scrambled when they pass through the mesh gateway wrapper anyway so we're not losing much. Another related fix is that we now use the endpoint health from the underlying service, not the health of the gateway (regardless of failover mode).
2019-08-05 18:30:35 +00:00
local DC=${3:-primary}
local NS=$4
test: for envoy integration tests, wait until 's2' is healthy in consul before interrogating envoy (#6108) When the envoy healthy panic threshold was explicitly disabled as part of L7 traffic management it changed how envoy decided to load balance to endpoints in a cluster. This only matters when envoy is in "panic mode" aka "when you have a bunch of unhealthy endpoints". Panic mode sends traffic to unhealthy instances in certain circumstances. Note: Prior to explicitly disabling the healthy panic threshold, the default value is 50%. What was happening is that the test harness was bringing up consul the sidecars, and the service instances all at once and sometimes the proxies wouldn't have time to be checked by consul to be labeled as 'passing' in the catalog before a round of EDS happened. The xDS server in consul effectively queries /v1/health/connect/s2 and gets 1 result, but that one result has a 'critical' check so the xDS server sends back that endpoint labeled as UNHEALTHY. Envoy sees that 100% of the endpoints in the cluster are unhealthy and would enter panic mode and still send traffic to s2. This is why the test suites PRIOR to disabling the healthy panic threshold worked. They were _incorrectly_ passing. When the healthy panic threshol is disabled, envoy never enters panic mode in this situation and thus the cluster has zero healthy endpoints so load balancing goes nowhere and the tests fail. Why does this only affect the test suites for envoy 1.8.0? My guess is that https://github.com/envoyproxy/envoy/pull/4442 was merged into the 1.9.x series and somehow that plays a role. This PR modifies the bats scripts to explicitly wait until the upstream sidecar is healthy as measured by /v1/health/connect/s2?passing BEFORE trying to interrogate envoy which should make the tests less racy.
2019-07-10 20:58:25 +00:00
GOT_COUNT=$(get_healthy_service_count "$SERVICE_NAME" "$DC" "$NS")
test: for envoy integration tests, wait until 's2' is healthy in consul before interrogating envoy (#6108) When the envoy healthy panic threshold was explicitly disabled as part of L7 traffic management it changed how envoy decided to load balance to endpoints in a cluster. This only matters when envoy is in "panic mode" aka "when you have a bunch of unhealthy endpoints". Panic mode sends traffic to unhealthy instances in certain circumstances. Note: Prior to explicitly disabling the healthy panic threshold, the default value is 50%. What was happening is that the test harness was bringing up consul the sidecars, and the service instances all at once and sometimes the proxies wouldn't have time to be checked by consul to be labeled as 'passing' in the catalog before a round of EDS happened. The xDS server in consul effectively queries /v1/health/connect/s2 and gets 1 result, but that one result has a 'critical' check so the xDS server sends back that endpoint labeled as UNHEALTHY. Envoy sees that 100% of the endpoints in the cluster are unhealthy and would enter panic mode and still send traffic to s2. This is why the test suites PRIOR to disabling the healthy panic threshold worked. They were _incorrectly_ passing. When the healthy panic threshol is disabled, envoy never enters panic mode in this situation and thus the cluster has zero healthy endpoints so load balancing goes nowhere and the tests fail. Why does this only affect the test suites for envoy 1.8.0? My guess is that https://github.com/envoyproxy/envoy/pull/4442 was merged into the 1.9.x series and somehow that plays a role. This PR modifies the bats scripts to explicitly wait until the upstream sidecar is healthy as measured by /v1/health/connect/s2?passing BEFORE trying to interrogate envoy which should make the tests less racy.
2019-07-10 20:58:25 +00:00
[ "$GOT_COUNT" -eq $EXPECT_COUNT ]
}
function assert_service_has_healthy_instances {
local SERVICE_NAME=$1
local EXPECT_COUNT=$2
connect: fix failover through a mesh gateway to a remote datacenter (#6259) Failover is pushed entirely down to the data plane by creating envoy clusters and putting each successive destination in a different load assignment priority band. For example this shows that normally requests go to 1.2.3.4:8080 but when that fails they go to 6.7.8.9:8080: - name: foo load_assignment: cluster_name: foo policy: overprovisioning_factor: 100000 endpoints: - priority: 0 lb_endpoints: - endpoint: address: socket_address: address: 1.2.3.4 port_value: 8080 - priority: 1 lb_endpoints: - endpoint: address: socket_address: address: 6.7.8.9 port_value: 8080 Mesh gateways route requests based solely on the SNI header tacked onto the TLS layer. Envoy currently only lets you configure the outbound SNI header at the cluster layer. If you try to failover through a mesh gateway you ideally would configure the SNI value per endpoint, but that's not possible in envoy today. This PR introduces a simpler way around the problem for now: 1. We identify any target of failover that will use mesh gateway mode local or remote and then further isolate any resolver node in the compiled discovery chain that has a failover destination set to one of those targets. 2. For each of these resolvers we will perform a small measurement of comparative healths of the endpoints that come back from the health API for the set of primary target and serial failover targets. We walk the list of targets in order and if any endpoint is healthy we return that target, otherwise we move on to the next target. 3. The CDS and EDS endpoints both perform the measurements in (2) for the affected resolver nodes. 4. For CDS this measurement selects which TLS SNI field to use for the cluster (note the cluster is always going to be named for the primary target) 5. For EDS this measurement selects which set of endpoints will populate the cluster. Priority tiered failover is ignored. One of the big downsides to this approach to failover is that the failover detection and correction is going to be controlled by consul rather than deferring that entirely to the data plane as with the prior version. This also means that we are bound to only failover using official health signals and cannot make use of data plane signals like outlier detection to affect failover. In this specific scenario the lack of data plane signals is ok because the effectiveness is already muted by the fact that the ultimate destination endpoints will have their data plane signals scrambled when they pass through the mesh gateway wrapper anyway so we're not losing much. Another related fix is that we now use the endpoint health from the underlying service, not the health of the gateway (regardless of failover mode).
2019-08-05 18:30:35 +00:00
local DC=${3:-primary}
local NS=$4
test: for envoy integration tests, wait until 's2' is healthy in consul before interrogating envoy (#6108) When the envoy healthy panic threshold was explicitly disabled as part of L7 traffic management it changed how envoy decided to load balance to endpoints in a cluster. This only matters when envoy is in "panic mode" aka "when you have a bunch of unhealthy endpoints". Panic mode sends traffic to unhealthy instances in certain circumstances. Note: Prior to explicitly disabling the healthy panic threshold, the default value is 50%. What was happening is that the test harness was bringing up consul the sidecars, and the service instances all at once and sometimes the proxies wouldn't have time to be checked by consul to be labeled as 'passing' in the catalog before a round of EDS happened. The xDS server in consul effectively queries /v1/health/connect/s2 and gets 1 result, but that one result has a 'critical' check so the xDS server sends back that endpoint labeled as UNHEALTHY. Envoy sees that 100% of the endpoints in the cluster are unhealthy and would enter panic mode and still send traffic to s2. This is why the test suites PRIOR to disabling the healthy panic threshold worked. They were _incorrectly_ passing. When the healthy panic threshol is disabled, envoy never enters panic mode in this situation and thus the cluster has zero healthy endpoints so load balancing goes nowhere and the tests fail. Why does this only affect the test suites for envoy 1.8.0? My guess is that https://github.com/envoyproxy/envoy/pull/4442 was merged into the 1.9.x series and somehow that plays a role. This PR modifies the bats scripts to explicitly wait until the upstream sidecar is healthy as measured by /v1/health/connect/s2?passing BEFORE trying to interrogate envoy which should make the tests less racy.
2019-07-10 20:58:25 +00:00
run retry_long assert_service_has_healthy_instances_once "$SERVICE_NAME" "$EXPECT_COUNT" "$DC" "$NS"
test: for envoy integration tests, wait until 's2' is healthy in consul before interrogating envoy (#6108) When the envoy healthy panic threshold was explicitly disabled as part of L7 traffic management it changed how envoy decided to load balance to endpoints in a cluster. This only matters when envoy is in "panic mode" aka "when you have a bunch of unhealthy endpoints". Panic mode sends traffic to unhealthy instances in certain circumstances. Note: Prior to explicitly disabling the healthy panic threshold, the default value is 50%. What was happening is that the test harness was bringing up consul the sidecars, and the service instances all at once and sometimes the proxies wouldn't have time to be checked by consul to be labeled as 'passing' in the catalog before a round of EDS happened. The xDS server in consul effectively queries /v1/health/connect/s2 and gets 1 result, but that one result has a 'critical' check so the xDS server sends back that endpoint labeled as UNHEALTHY. Envoy sees that 100% of the endpoints in the cluster are unhealthy and would enter panic mode and still send traffic to s2. This is why the test suites PRIOR to disabling the healthy panic threshold worked. They were _incorrectly_ passing. When the healthy panic threshol is disabled, envoy never enters panic mode in this situation and thus the cluster has zero healthy endpoints so load balancing goes nowhere and the tests fail. Why does this only affect the test suites for envoy 1.8.0? My guess is that https://github.com/envoyproxy/envoy/pull/4442 was merged into the 1.9.x series and somehow that plays a role. This PR modifies the bats scripts to explicitly wait until the upstream sidecar is healthy as measured by /v1/health/connect/s2?passing BEFORE trying to interrogate envoy which should make the tests less racy.
2019-07-10 20:58:25 +00:00
[ "$status" -eq 0 ]
}
function check_intention {
local SOURCE=$1
local DESTINATION=$2
curl -s -f "localhost:8500/v1/connect/intentions/check?source=${SOURCE}&destination=${DESTINATION}" | jq ".Allowed"
}
function assert_intention_allowed {
local SOURCE=$1
local DESTINATION=$2
[ "$(check_intention "${SOURCE}" "${DESTINATION}")" == "true" ]
}
function assert_intention_denied {
local SOURCE=$1
local DESTINATION=$2
[ "$(check_intention "${SOURCE}" "${DESTINATION}")" == "false" ]
}
function docker_consul {
local DC=$1
shift 1
docker run -i --rm --network container:envoy_consul-${DC}_1 consul-dev "$@"
}
function docker_wget {
local DC=$1
shift 1
docker run -ti --rm --network container:envoy_consul-${DC}_1 alpine:3.9 wget "$@"
}
function docker_curl {
local DC=$1
shift 1
docker run -ti --rm --network container:envoy_consul-${DC}_1 --entrypoint curl consul-dev "$@"
}
function docker_exec {
if ! docker exec -i "$@"
then
echo "Failed to execute: docker exec -i $@" 1>&2
return 1
fi
}
function docker_consul_exec {
local DC=$1
shift 1
docker_exec envoy_consul-${DC}_1 "$@"
}
function get_envoy_pid {
local BOOTSTRAP_NAME=$1
local DC=${2:-primary}
run ps aux
[ "$status" == 0 ]
echo "$output" 1>&2
PID="$(echo "$output" | grep "envoy -c /workdir/$DC/envoy/${BOOTSTRAP_NAME}-bootstrap.json" | awk '{print $1}')"
[ -n "$PID" ]
echo "$PID"
}
function kill_envoy {
local BOOTSTRAP_NAME=$1
local DC=${2:-primary}
PID="$(get_envoy_pid $BOOTSTRAP_NAME "$DC")"
echo "PID = $PID"
kill -TERM $PID
}
function must_match_in_statsd_logs {
local DC=${2:-primary}
run cat /workdir/${DC}/statsd/statsd.log
echo "$output"
COUNT=$( echo "$output" | grep -Ec $1 )
echo "COUNT of '$1' matches: $COUNT"
[ "$status" == 0 ]
[ "$COUNT" -gt "0" ]
}
function must_match_in_prometheus_response {
run curl -f -s $1/metrics
COUNT=$( echo "$output" | grep -Ec $2 )
echo "OUTPUT head -n 10"
echo "$output" | head -n 10
echo "COUNT of '$2' matches: $COUNT"
[ "$status" == 0 ]
[ "$COUNT" -gt "0" ]
}
function must_match_in_stats_proxy_response {
run curl -f -s $1/$2
COUNT=$( echo "$output" | grep -Ec $3 )
echo "OUTPUT head -n 10"
echo "$output" | head -n 10
echo "COUNT of '$3' matches: $COUNT"
[ "$status" == 0 ]
[ "$COUNT" -gt "0" ]
}
# must_fail_tcp_connection checks that a request made through an upstream fails,
# probably due to authz being denied if all other tests passed already. Although
# we are using curl, this only works as expected for TCP upstreams as we are
# checking TCP-level errors. HTTP upstreams will return a valid 503 generated by
# Envoy rather than a connection-level error.
function must_fail_tcp_connection {
# Attempt to curl through upstream
run curl -s -v -f -d hello $1
echo "OUTPUT $output"
# Should fail during handshake and return "got nothing" error
[ "$status" == "52" ]
# Verbose output should enclude empty reply
echo "$output" | grep 'Empty reply from server'
}
# must_fail_http_connection see must_fail_tcp_connection but this expects Envoy
# to generate a 503 response since the upstreams have refused connection.
function must_fail_http_connection {
# Attempt to curl through upstream
run curl -s -i -d hello $1
echo "OUTPUT $output"
# Should fail request with 503
echo "$output" | grep '503 Service Unavailable'
}
function gen_envoy_bootstrap {
SERVICE=$1
ADMIN_PORT=$2
DC=${3:-primary}
IS_MGW=${4:-0}
EXTRA_ENVOY_BS_ARGS="${5-}"
PROXY_ID="$SERVICE"
if ! is_set "$IS_MGW"
then
PROXY_ID="$SERVICE-sidecar-proxy"
fi
if output=$(docker_consul "$DC" connect envoy -bootstrap \
-proxy-id $PROXY_ID \
-envoy-version "$ENVOY_VERSION" \
-admin-bind 0.0.0.0:$ADMIN_PORT ${EXTRA_ENVOY_BS_ARGS} 2>&1); then
# All OK, write config to file
echo "$output" > workdir/${DC}/envoy/$SERVICE-bootstrap.json
else
status=$?
# Command failed, instead of swallowing error (printed on stdout by docker
# it seems) by writing it to file, echo it
echo "$output"
return $status
fi
}
function read_config_entry {
local KIND=$1
local NAME=$2
local DC=${3:-primary}
docker_consul "$DC" config read -kind $KIND -name $NAME
}
function wait_for_config_entry {
retry_default read_config_entry "$@" >/dev/null
}
function delete_config_entry {
local KIND=$1
local NAME=$2
retry_default curl -sL -XDELETE "http://127.0.0.1:8500/v1/config/${KIND}/${NAME}"
}
function list_intentions {
curl -s -f "http://localhost:8500/v1/connect/intentions"
}
function get_intention_target_name {
awk -F / '{ if ( NF == 1 ) { print $0 } else { print $2 }}'
}
function get_intention_target_namespace {
awk -F / '{ if ( NF != 1 ) { print $1 } }'
}
function get_intention_by_targets {
local SOURCE=$1
local DESTINATION=$2
local SOURCE_NS=$(get_intention_target_namespace <<< "${SOURCE}")
local SOURCE_NAME=$(get_intention_target_name <<< "${SOURCE}")
local DESTINATION_NS=$(get_intention_target_namespace <<< "${DESTINATION}")
local DESTINATION_NAME=$(get_intention_target_name <<< "${DESTINATION}")
existing=$(list_intentions | jq ".[] | select(.SourceNS == \"$SOURCE_NS\" and .SourceName == \"$SOURCE_NAME\" and .DestinationNS == \"$DESTINATION_NS\" and .DestinationName == \"$DESTINATION_NAME\")")
if test -z "$existing"
then
return 1
fi
echo "$existing"
return 0
}
function update_intention {
local SOURCE=$1
local DESTINATION=$2
local ACTION=$3
intention=$(get_intention_by_targets "${SOURCE}" "${DESTINATION}")
if test $? -ne 0
then
return 1
fi
id=$(jq -r .ID <<< "${intention}")
updated=$(jq ".Action = \"$ACTION\"" <<< "${intention}")
curl -s -X PUT "http://localhost:8500/v1/connect/intentions/${id}" -d "${updated}"
return $?
}
function wait_for_agent_service_register {
local SERVICE_ID=$1
local DC=${2:-primary}
retry_default docker_curl "$DC" -sLf "http://127.0.0.1:8500/v1/agent/service/${SERVICE_ID}" >/dev/null
}
function set_ttl_check_state {
local CHECK_ID=$1
local CHECK_STATE=$2
local DC=${3:-primary}
case "$CHECK_STATE" in
pass)
;;
warn)
;;
fail)
;;
*)
echo "invalid ttl check state '${CHECK_STATE}'" >&2
return 1
esac
retry_default docker_curl "${DC}" -sL -XPUT "http://localhost:8500/v1/agent/check/warn/${CHECK_ID}"
}
function get_upstream_fortio_name {
run retry_default curl -v -s -f localhost:5000/debug?env=dump
[ "$status" == 0 ]
echo "$output" | grep -E "^FORTIO_NAME="
}
function assert_expected_fortio_name {
local EXPECT_NAME=$1
GOT=$(get_upstream_fortio_name)
connect: fix failover through a mesh gateway to a remote datacenter (#6259) Failover is pushed entirely down to the data plane by creating envoy clusters and putting each successive destination in a different load assignment priority band. For example this shows that normally requests go to 1.2.3.4:8080 but when that fails they go to 6.7.8.9:8080: - name: foo load_assignment: cluster_name: foo policy: overprovisioning_factor: 100000 endpoints: - priority: 0 lb_endpoints: - endpoint: address: socket_address: address: 1.2.3.4 port_value: 8080 - priority: 1 lb_endpoints: - endpoint: address: socket_address: address: 6.7.8.9 port_value: 8080 Mesh gateways route requests based solely on the SNI header tacked onto the TLS layer. Envoy currently only lets you configure the outbound SNI header at the cluster layer. If you try to failover through a mesh gateway you ideally would configure the SNI value per endpoint, but that's not possible in envoy today. This PR introduces a simpler way around the problem for now: 1. We identify any target of failover that will use mesh gateway mode local or remote and then further isolate any resolver node in the compiled discovery chain that has a failover destination set to one of those targets. 2. For each of these resolvers we will perform a small measurement of comparative healths of the endpoints that come back from the health API for the set of primary target and serial failover targets. We walk the list of targets in order and if any endpoint is healthy we return that target, otherwise we move on to the next target. 3. The CDS and EDS endpoints both perform the measurements in (2) for the affected resolver nodes. 4. For CDS this measurement selects which TLS SNI field to use for the cluster (note the cluster is always going to be named for the primary target) 5. For EDS this measurement selects which set of endpoints will populate the cluster. Priority tiered failover is ignored. One of the big downsides to this approach to failover is that the failover detection and correction is going to be controlled by consul rather than deferring that entirely to the data plane as with the prior version. This also means that we are bound to only failover using official health signals and cannot make use of data plane signals like outlier detection to affect failover. In this specific scenario the lack of data plane signals is ok because the effectiveness is already muted by the fact that the ultimate destination endpoints will have their data plane signals scrambled when they pass through the mesh gateway wrapper anyway so we're not losing much. Another related fix is that we now use the endpoint health from the underlying service, not the health of the gateway (regardless of failover mode).
2019-08-05 18:30:35 +00:00
if [ "$GOT" != "FORTIO_NAME=${EXPECT_NAME}" ]; then
echo "expected name: $EXPECT_NAME, actual name: $GOT" 1>&2
return 1
fi
}