From 24f140f9ed1a2180541c0c7cebf7572c5155fe27 Mon Sep 17 00:00:00 2001
From: Tom Clegg <tom@curii.com>
Date: Mon, 6 Dec 2021 19:32:34 -0500
Subject: [PATCH] 18547: Error out if two volumes return the same non-empty
 DeviceID.

Arvados-DCO-1.1-Signed-off-by: Tom Clegg <tom@curii.com>
---
 services/keep-balance/balance.go          | 24 +++++++++++++++++++++
 services/keep-balance/balance_run_test.go | 26 +++++++++++++++++++++++
 2 files changed, 50 insertions(+)

diff --git a/services/keep-balance/balance.go b/services/keep-balance/balance.go
index fa01d512bc..eb6f580f43 100644
--- a/services/keep-balance/balance.go
+++ b/services/keep-balance/balance.go
@@ -8,6 +8,7 @@ import (
 	"bytes"
 	"context"
 	"crypto/md5"
+	"errors"
 	"fmt"
 	"io"
 	"io/ioutil"
@@ -266,6 +267,29 @@ func (bal *Balancer) CheckSanityEarly(c *arvados.Client) error {
 		}
 	}
 
+	mountProblem := false
+	type deviceMount struct {
+		srv *KeepService
+		mnt *KeepMount
+	}
+	deviceMounted := map[string]deviceMount{} // DeviceID -> mount
+	for _, srv := range bal.KeepServices {
+		for _, mnt := range srv.mounts {
+			if first, dup := deviceMounted[mnt.DeviceID]; dup && first.mnt.UUID != mnt.UUID && mnt.DeviceID != "" {
+				bal.logf("config error: device %s is mounted with multiple volume UUIDs: %s on %s, and %s on %s",
+					mnt.DeviceID,
+					first.mnt.UUID, first.srv,
+					mnt.UUID, srv)
+				mountProblem = true
+				continue
+			}
+			deviceMounted[mnt.DeviceID] = deviceMount{srv, mnt}
+		}
+	}
+	if mountProblem {
+		return errors.New("cannot continue with config errors (see above)")
+	}
+
 	var checkPage arvados.CollectionList
 	if err = c.RequestAndDecode(&checkPage, "GET", "arvados/v1/collections", nil, arvados.ResourceListParams{
 		Limit:              new(int),
diff --git a/services/keep-balance/balance_run_test.go b/services/keep-balance/balance_run_test.go
index 4e2c6803ca..0d1b6b5912 100644
--- a/services/keep-balance/balance_run_test.go
+++ b/services/keep-balance/balance_run_test.go
@@ -397,6 +397,32 @@ func (s *runSuite) TestRefuseNonAdmin(c *check.C) {
 	c.Check(pullReqs.Count(), check.Equals, 0)
 }
 
+func (s *runSuite) TestRefuseSameDeviceDifferentVolumes(c *check.C) {
+	opts := RunOptions{
+		CommitPulls: true,
+		CommitTrash: true,
+		Logger:      ctxlog.TestLogger(c),
+	}
+	s.stub.serveCurrentUserAdmin()
+	s.stub.serveZeroCollections()
+	s.stub.serveKeepServices(stubServices)
+	s.stub.mux.HandleFunc("/mounts", func(w http.ResponseWriter, r *http.Request) {
+		hostid := r.Host[:5] // "keep0.zzzzz.arvadosapi.com:25107" => "keep0"
+		json.NewEncoder(w).Encode([]arvados.KeepMount{{
+			UUID:           "zzzzz-ivpuk-0000000000" + hostid,
+			DeviceID:       "keep0-vol0",
+			StorageClasses: map[string]bool{"default": true},
+		}})
+	})
+	trashReqs := s.stub.serveKeepstoreTrash()
+	pullReqs := s.stub.serveKeepstorePull()
+	srv := s.newServer(&opts)
+	_, err := srv.runOnce()
+	c.Check(err, check.ErrorMatches, "cannot continue with config errors.*")
+	c.Check(trashReqs.Count(), check.Equals, 0)
+	c.Check(pullReqs.Count(), check.Equals, 0)
+}
+
 func (s *runSuite) TestWriteLostBlocks(c *check.C) {
 	lostf, err := ioutil.TempFile("", "keep-balance-lost-blocks-test-")
 	c.Assert(err, check.IsNil)
-- 
2.30.2