Merge pull request #30 from fly-apps/failover-verification

Failover verification

Author: davissp14

Dockerfile

@@ -7,6 +7,8 @@ WORKDIR /go/src/github.com/fly-examples/fly-postgres
 COPY . .
 
 RUN CGO_ENABLED=0 GOOS=linux go build -v -o /fly/bin/event_handler ./cmd/event_handler
+RUN CGO_ENABLED=0 GOOS=linux go build -v -o /fly/bin/failover_validation ./cmd/failover_validation
+
 RUN CGO_ENABLED=0 GOOS=linux go build -v -o /fly/bin/standby_cleaner ./cmd/standby_cleaner
 RUN CGO_ENABLED=0 GOOS=linux go build -v -o /fly/bin/start ./cmd/start
 COPY ./bin/* /fly/bin/

cmd/failover_validation/main.go

@@ -0,0 +1,24 @@
+package main
+
+import (
+	"flag"
+	"fmt"
+	"os"
+)
+
+func main() {
+	visibleNodes := flag.Int("visible-nodes", 0, "Total visible nodes from the perspective of the proposed leader")
+	totalNodes := flag.Int("total-nodes", 0, "The total number of nodes registered")
+	flag.Parse()
+
+	// TODO - This will ultimately remove HA from a 2-node cluster setup.
+	// This will be the case until we come up with a strategy to differentiate
+	// between a down node and a network partition.
+
+	if *visibleNodes == 0 || *visibleNodes < (*totalNodes/2+1) {
+		fmt.Printf("Unable to perform failover as quorum can not be met. Total nodes: %d, Visible nodes: %d\n", *totalNodes, *visibleNodes)
+		os.Exit(1)
+	}
+
+	os.Exit(0)
+}

pkg/flypg/repmgr.go

@@ -135,6 +135,7 @@ func (r *RepMgr) setDefaults() {
 		"event_notifications":          "'repmgrd_failover_promote,standby_promote,standby_follow'",
 		"location":                     r.Region,
 		"primary_visibility_consensus": true,
+		"failover_validation_command":  fmt.Sprintf("'/usr/local/bin/failover_validation -visible-nodes %%v -total-nodes %%t'"),
 	}
 
 	if !r.eligiblePrimary() {