From 92802da2ddf97e305fb310dc71a30ccd55a7c0e9 Mon Sep 17 00:00:00 2001
From: Brian Picciano <me@mediocregopher.com>
Date: Tue, 7 Jan 2025 15:40:50 +0100
Subject: [PATCH] Use GLM for garage layout management

---
 go/bootstrap/hosts.go                         |  10 +-
 go/daemon/children/children.go                | 101 ++++-
 go/daemon/children/garage.go                  |  24 +-
 go/daemon/daecommon/config.go                 |  16 -
 go/daemon/network/bootstrap.go                |  34 +-
 go/daemon/network/garage.go                   |  67 ++--
 go/daemon/network/network.go                  | 353 +++++++++++++-----
 go/daemon/network/network_it_test.go          | 259 ++++++-------
 go/daemon/network/network_it_util_test.go     |  85 +++--
 go/garage/admin_client.go                     |   1 +
 go/garage/garagesrv/garagesrv.go              | 136 +++----
 go/garage/peer.go                             |  34 +-
 .../validate-non-conflicting-ports-or-dirs.md |  10 +
 tasks/misc/storage-allocation-status.md       |  10 +
 14 files changed, 670 insertions(+), 470 deletions(-)
 create mode 100644 tasks/bugs/validate-non-conflicting-ports-or-dirs.md
 create mode 100644 tasks/misc/storage-allocation-status.md

diff --git a/go/bootstrap/hosts.go b/go/bootstrap/hosts.go
index f0ded15..af6b3a1 100644
--- a/go/bootstrap/hosts.go
+++ b/go/bootstrap/hosts.go
@@ -124,10 +124,12 @@ func (h Host) GarageNodes() []garage.RemoteNode {
 	var nodes []garage.RemoteNode
 	for _, instance := range h.Garage.Instances {
 		nodes = append(nodes, garage.RemoteNode{
-			ID:        instance.ID,
-			IP:        h.IP().String(),
-			RPCPort:   instance.RPCPort,
-			S3APIPort: instance.S3APIPort,
+			Node: garage.Node{
+				IP:        h.IP().String(),
+				RPCPort:   instance.RPCPort,
+				S3APIPort: instance.S3APIPort,
+			},
+			ID: instance.ID,
 		})
 	}
 	return nodes
diff --git a/go/daemon/children/children.go b/go/daemon/children/children.go
index f782bad..36d716e 100644
--- a/go/daemon/children/children.go
+++ b/go/daemon/children/children.go
@@ -17,6 +17,7 @@ import (
 	"isle/bootstrap"
 	"isle/daemon/daecommon"
 	"isle/garage"
+	"isle/garage/garagesrv"
 	"isle/secrets"
 	"isle/toolkit"
 )
@@ -33,12 +34,28 @@ type Opts struct {
 	// GarageNewCluster should be true if the garage instances being started
 	// are the first instances in a cluster which is being created.
 	GarageNewCluster bool
+
+	// GarageBootstrapPeers will be used as the set of peers each garage
+	// instance should use to find the rest of the garage cluster.
+	//
+	// Defaults to peer information contained in the bootstrap hosts.
+	GarageBootstrapPeers []garage.RemoteNode
+
+	// DEPRECATED can be used to manually set the db engine used by garage for
+	// new allocations. If not given then garagesrv.DBEngineSqlite will be used
+	// for new allocations.
+	GarageDefaultDBEngine garagesrv.DBEngine
 }
 
 func (o *Opts) withDefaults() *Opts {
 	if o == nil {
 		o = new(Opts)
 	}
+
+	if o.GarageDefaultDBEngine == "" {
+		o.GarageDefaultDBEngine = garageDefaultDBEngine
+	}
+
 	return o
 }
 
@@ -53,6 +70,7 @@ type Children struct {
 	runtimeDir        toolkit.Dir
 	garageAdminToken  string
 	nebulaDeviceNamer *NebulaDeviceNamer
+	opts              *Opts
 
 	garageRPCSecret string
 
@@ -91,6 +109,7 @@ func New(
 		runtimeDir:        runtimeDir,
 		garageAdminToken:  garageAdminToken,
 		nebulaDeviceNamer: nebulaDeviceNamer,
+		opts:              opts,
 		garageRPCSecret:   garageRPCSecret,
 	}
 
@@ -125,6 +144,11 @@ func New(
 		return nil, fmt.Errorf("starting dnsmasq: %w", err)
 	}
 
+	garageBootstrapPeers := opts.GarageBootstrapPeers
+	if garageBootstrapPeers == nil {
+		garageBootstrapPeers = hostBootstrap.GarageNodes()
+	}
+
 	if c.garageProcs, err = garagePmuxProcs(
 		ctx,
 		c.logger,
@@ -134,6 +158,8 @@ func New(
 		networkConfig,
 		garageAdminToken,
 		hostBootstrap,
+		garageBootstrapPeers,
+		c.opts.GarageDefaultDBEngine,
 	); err != nil {
 		logger.Warn(ctx, "Failed to start garage processes, shutting down child processes", err)
 		c.Shutdown()
@@ -209,24 +235,28 @@ func (c *Children) reloadGarage(
 	networkConfig daecommon.NetworkConfig,
 	hostBootstrap bootstrap.Bootstrap,
 ) error {
-	allocs := networkConfig.Storage.Allocations
-	if len(allocs) == 0 {
-		return nil
-	}
 
-	thisHost := hostBootstrap.ThisHost()
+	var (
+		allocs     = networkConfig.Storage.Allocations
+		thisHost   = hostBootstrap.ThisHost()
+		anyStarted bool
+		allocsM    = map[daecommon.ConfigStorageAllocation]struct{}{}
+	)
 
-	var anyCreated bool
 	for _, alloc := range allocs {
+		allocsM[alloc] = struct{}{}
+
 		var (
 			procName = garagePmuxProcName(alloc)
 			ctx      = mctx.Annotate(
-				ctx,
+				mctx.WithAnnotator(ctx, alloc),
 				"garageProcName", procName,
-				"garageDataPath", alloc.DataPath,
 			)
 		)
 
+		// Rewrite the child config always, even if we don't always restart it.
+		// If nothing else this will capture any changes to the bootstrap nodes,
+		// which will be useful if garage gets restarted for any reason.
 		childConfigPath, err := garageWriteChildConfig(
 			ctx,
 			c.logger,
@@ -234,20 +264,31 @@ func (c *Children) reloadGarage(
 			c.runtimeDir.Path,
 			c.garageAdminToken,
 			hostBootstrap,
+			hostBootstrap.GarageNodes(),
 			alloc,
+			garageDefaultDBEngine,
 		)
 		if err != nil {
 			return fmt.Errorf("writing child config file for alloc %+v: %w", alloc, err)
 		}
 
-		if _, ok := c.garageProcs[procName]; ok {
-			c.logger.Info(ctx, "Garage instance already exists")
+		if proc, ok := c.garageProcs[procName]; ok {
+			if proc.alloc == alloc {
+				c.logger.Info(ctx, "No changes to storage allocation, leaving garage process as-is")
+				continue
+			}
+
+			c.logger.Info(ctx, "Storage allocation modified, restarting garage process")
+			proc.Restart()
+			anyStarted = true
+
+			proc.alloc = alloc
+			c.garageProcs[procName] = proc
+
 			continue
 		}
 
-		anyCreated = true
-
-		c.logger.Info(ctx, "Garage config has been added, creating process")
+		c.logger.Info(ctx, "New storage allocation, creating garage process")
 		c.garageProcs[procName] = garageProc{
 			Process: garagePmuxProc(
 				ctx, c.logger, c.binDirPath, procName, childConfigPath,
@@ -255,9 +296,26 @@ func (c *Children) reloadGarage(
 			alloc:     alloc,
 			adminAddr: garageAllocAdminAddr(thisHost, alloc),
 		}
+
+		anyStarted = true
 	}
 
-	if anyCreated {
+	for procName, proc := range c.garageProcs {
+		if _, ok := allocsM[proc.alloc]; ok {
+			continue
+		}
+
+		ctx := mctx.Annotate(
+			mctx.WithAnnotator(ctx, proc.alloc),
+			"garageProcName", procName,
+		)
+
+		c.logger.Info(ctx, "Storage allocation removed, stopping garage process")
+		proc.Stop()
+		delete(c.garageProcs, procName)
+	}
+
+	if anyStarted {
 		if err := waitForGarage(
 			ctx,
 			c.logger,
@@ -296,6 +354,21 @@ func (c *Children) Reload(
 	return errors.Join(errs...)
 }
 
+// ActiveStorageAllocations returns the storage allocations which currently have
+// active garage instances.
+func (c *Children) ActiveStorageAllocations() []daecommon.ConfigStorageAllocation {
+	allocs := make([]daecommon.ConfigStorageAllocation, 0, len(c.garageProcs))
+	for _, proc := range c.garageProcs {
+		allocs = append(allocs, proc.alloc)
+	}
+
+	slices.SortFunc(allocs, func(a, b daecommon.ConfigStorageAllocation) int {
+		return cmp.Compare(a.RPCPort, b.RPCPort)
+	})
+
+	return allocs
+}
+
 // GarageAdminClient returns an admin client for an active local garage process,
 // or false if there are no garage processes.
 func (c *Children) GarageAdminClient() (*garage.AdminClient, bool) {
diff --git a/go/daemon/children/garage.go b/go/daemon/children/garage.go
index fcc00b0..4e3286a 100644
--- a/go/daemon/children/garage.go
+++ b/go/daemon/children/garage.go
@@ -17,6 +17,10 @@ import (
 	"dev.mediocregopher.com/mediocre-go-lib.git/mlog"
 )
 
+const (
+	garageDefaultDBEngine = garagesrv.DBEngineSqlite
+)
+
 func garageAdminClientLogger(logger *mlog.Logger) *mlog.Logger {
 	return logger.WithNamespace("garageAdminClient")
 }
@@ -42,9 +46,8 @@ func waitForGarage(
 		)
 
 		ctx := mctx.Annotate(
-			ctx,
+			mctx.WithAnnotator(ctx, proc.alloc),
 			"garageAdminAddr", proc.adminAddr,
-			"garageDataPath", proc.alloc.DataPath,
 		)
 
 		logger.Info(ctx, "Waiting for garage instance to be healthy")
@@ -70,17 +73,16 @@ func garageWriteChildConfig(
 	logger *mlog.Logger,
 	rpcSecret, runtimeDirPath, adminToken string,
 	hostBootstrap bootstrap.Bootstrap,
+	bootstrapPeers []garage.RemoteNode,
 	alloc daecommon.ConfigStorageAllocation,
+	defaultDBEngine garagesrv.DBEngine,
 ) (
 	string, error,
 ) {
 	var (
 		thisHost = hostBootstrap.ThisHost()
-		id       = daecommon.BootstrapGarageHostForAlloc(thisHost, alloc).ID
-
-		node = garage.LocalNode{
-			RemoteNode: garage.RemoteNode{
-				ID:        id,
+		node     = garage.LocalNode{
+			Node: garage.Node{
 				IP:        thisHost.IP().String(),
 				RPCPort:   alloc.RPCPort,
 				S3APIPort: alloc.S3APIPort,
@@ -93,7 +95,7 @@ func garageWriteChildConfig(
 		)
 	)
 
-	dbEngine, err := garagesrv.GetDBEngine(alloc.MetaPath)
+	dbEngine, err := garagesrv.GetDBEngine(alloc.MetaPath, defaultDBEngine)
 	if err != nil {
 		return "", fmt.Errorf("getting alloc db engine: %w", err)
 	}
@@ -111,7 +113,7 @@ func garageWriteChildConfig(
 			DBEngine:   dbEngine,
 
 			LocalNode:      node,
-			BootstrapPeers: hostBootstrap.GarageNodes(),
+			BootstrapPeers: bootstrapPeers,
 		},
 	)
 
@@ -152,6 +154,8 @@ func garagePmuxProcs(
 	networkConfig daecommon.NetworkConfig,
 	adminToken string,
 	hostBootstrap bootstrap.Bootstrap,
+	bootstrapPeers []garage.RemoteNode,
+	defaultDBEngine garagesrv.DBEngine,
 ) (
 	map[string]garageProc, error,
 ) {
@@ -171,7 +175,9 @@ func garagePmuxProcs(
 			logger,
 			rpcSecret, runtimeDirPath, adminToken,
 			hostBootstrap,
+			bootstrapPeers,
 			alloc,
+			defaultDBEngine,
 		)
 		if err != nil {
 			return nil, fmt.Errorf("writing child config file for alloc %+v: %w", alloc, err)
diff --git a/go/daemon/daecommon/config.go b/go/daemon/daecommon/config.go
index 7ded765..c5900e0 100644
--- a/go/daemon/daecommon/config.go
+++ b/go/daemon/daecommon/config.go
@@ -5,7 +5,6 @@ import (
 	"encoding/json"
 	"fmt"
 	"io"
-	"isle/bootstrap"
 	"isle/toolkit"
 	"isle/yamlutil"
 	"net"
@@ -325,18 +324,3 @@ func LoadConfig(userConfigPath string) (Config, error) {
 	err := yamlutil.LoadYamlFile(&config, userConfigPath)
 	return config, err
 }
-
-// BootstrapGarageHostForAlloc returns the bootstrap.GarageHostInstance which
-// corresponds with the given alloc from the daemon config. This will panic if
-// no associated instance can be found.
-func BootstrapGarageHostForAlloc(
-	host bootstrap.Host, alloc ConfigStorageAllocation,
-) bootstrap.GarageHostInstance {
-	for _, inst := range host.Garage.Instances {
-		if inst.RPCPort == alloc.RPCPort {
-			return inst
-		}
-	}
-
-	panic(fmt.Sprintf("could not find alloc %+v in the bootstrap data", alloc))
-}
diff --git a/go/daemon/network/bootstrap.go b/go/daemon/network/bootstrap.go
index 9b51eb1..02f983a 100644
--- a/go/daemon/network/bootstrap.go
+++ b/go/daemon/network/bootstrap.go
@@ -63,28 +63,22 @@ func applyNetworkConfigToBootstrap(
 		},
 	}
 
-	if allocs := networkConfig.Storage.Allocations; len(allocs) > 0 {
-
-		for i, alloc := range allocs {
-
-			id, rpcPort, err := garagesrv.InitAlloc(alloc.MetaPath, alloc.RPCPort)
-			if err != nil {
-				return bootstrap.Bootstrap{}, fmt.Errorf(
-					"initializing alloc at %q: %w", alloc.MetaPath, err,
-				)
-			}
-
-			host.Garage.Instances = append(
-				host.Garage.Instances,
-				bootstrap.GarageHostInstance{
-					ID:        id,
-					RPCPort:   rpcPort,
-					S3APIPort: alloc.S3APIPort,
-				},
+	for _, alloc := range networkConfig.Storage.Allocations {
+		id, err := garagesrv.LoadAllocID(alloc.MetaPath)
+		if err != nil {
+			return bootstrap.Bootstrap{}, fmt.Errorf(
+				"getting ID of alloc at %q: %w", alloc.MetaPath, err,
 			)
-
-			allocs[i].RPCPort = rpcPort
 		}
+
+		host.Garage.Instances = append(
+			host.Garage.Instances,
+			bootstrap.GarageHostInstance{
+				ID:        id,
+				RPCPort:   alloc.RPCPort,
+				S3APIPort: alloc.S3APIPort,
+			},
+		)
 	}
 
 	hostBootstrap.Hosts = maps.Clone(hostBootstrap.Hosts)
diff --git a/go/daemon/network/garage.go b/go/daemon/network/garage.go
index 220c8f6..2241680 100644
--- a/go/daemon/network/garage.go
+++ b/go/daemon/network/garage.go
@@ -9,9 +9,11 @@ import (
 	"isle/bootstrap"
 	"isle/daemon/daecommon"
 	"isle/garage"
+	"isle/garage/garagesrv"
 	"isle/nebula"
 	"isle/secrets"
 	"isle/toolkit"
+	"net/netip"
 	"path/filepath"
 	"time"
 
@@ -51,27 +53,47 @@ func getGarageClientParams(
 	}, nil
 }
 
-func garageApplyLayout(
-	ctx context.Context,
-	logger *mlog.Logger,
-	networkConfig daecommon.NetworkConfig,
-	adminClient *garage.AdminClient,
-	prevHost, currHost bootstrap.Host,
-) error {
-	var (
-		hostName = currHost.Name
-		allocs   = networkConfig.Storage.Allocations
-		roles    = make([]garage.Role, len(allocs))
-		roleIDs  = map[string]struct{}{}
+func garageInitAllocs(
+	hostIP netip.Addr,
+	allocs []daecommon.ConfigStorageAllocation,
+) (
+	[]garage.RemoteNode, error,
+) {
+	peers := make([]garage.RemoteNode, len(allocs))
+	for i, alloc := range allocs {
+		id, err := garagesrv.InitAlloc(alloc.MetaPath)
+		if err != nil {
+			return nil, fmt.Errorf("initializing alloc %+v: %w", alloc, err)
+		}
 
-		idsToRemove = make([]string, 0, len(prevHost.Garage.Instances))
+		peers[i] = garage.RemoteNode{
+			Node: garage.Node{
+				IP:        hostIP.String(),
+				RPCPort:   alloc.RPCPort,
+				S3APIPort: alloc.S3APIPort,
+			},
+			ID: id,
+		}
+	}
+
+	return peers, nil
+}
+
+func garageAllocsToRoles(
+	host bootstrap.Host, allocs []daecommon.ConfigStorageAllocation,
+) (
+	[]garage.Role, error,
+) {
+	var (
+		hostName = host.Name
+		roles    = make([]garage.Role, len(allocs))
 	)
 
-	defer adminClient.Close()
-
 	for i, alloc := range allocs {
-		id := daecommon.BootstrapGarageHostForAlloc(currHost, alloc).ID
-		roleIDs[id] = struct{}{}
+		id, err := garagesrv.LoadAllocID(alloc.MetaPath)
+		if err != nil {
+			return nil, fmt.Errorf("getting ID of alloc %+v: %w", alloc, err)
+		}
 
 		roles[i] = garage.Role{
 			ID:       id,
@@ -81,21 +103,12 @@ func garageApplyLayout(
 		}
 	}
 
-	for _, prevInst := range prevHost.Garage.Instances {
-		if _, ok := roleIDs[prevInst.ID]; !ok {
-			idsToRemove = append(idsToRemove, prevInst.ID)
-		}
-	}
-
-	return adminClient.ApplyLayout(ctx, roles, idsToRemove)
+	return roles, nil
 }
 
 func garageInitializeGlobalBucket(
 	ctx context.Context,
-	logger *mlog.Logger,
-	networkConfig daecommon.NetworkConfig,
 	adminClient *garage.AdminClient,
-	host bootstrap.Host,
 ) (
 	garage.S3APICredentials, error,
 ) {
diff --git a/go/daemon/network/network.go b/go/daemon/network/network.go
index 1114463..57c51a5 100644
--- a/go/daemon/network/network.go
+++ b/go/daemon/network/network.go
@@ -14,7 +14,9 @@ import (
 	"isle/bootstrap"
 	"isle/daemon/children"
 	"isle/daemon/daecommon"
+	"isle/daemon/network/glm"
 	"isle/garage"
+	"isle/garage/garagesrv"
 	"isle/jsonutil"
 	"isle/nebula"
 	"isle/secrets"
@@ -163,6 +165,9 @@ type Opts struct {
 
 	// testBlocker is used by tests to set blockpoints.
 	testBlocker *toolkit.TestBlocker
+
+	// DEPRECATED See corresponding field in [children.Opts]
+	garageDefaultDBEngine garagesrv.DBEngine
 }
 
 func (o *Opts) withDefaults() *Opts {
@@ -187,7 +192,8 @@ type network struct {
 
 	opts *Opts
 
-	secretsStore secrets.Store
+	secretsStore    secrets.Store
+	garageLayoutMgr glm.GarageLayoutManager
 
 	l             sync.RWMutex
 	children      *children.Children
@@ -218,13 +224,15 @@ func newNetwork(
 	ctx, cancel := context.WithCancel(context.WithoutCancel(ctx))
 
 	var (
-		n = &network{
+		ip = currBootstrap.ThisHost().IP()
+		n  = &network{
 			logger:            logger,
 			envBinDirPath:     envBinDirPath,
 			nebulaDeviceNamer: nebulaDeviceNamer,
 			stateDir:          stateDir,
 			runtimeDir:        runtimeDir,
 			opts:              opts.withDefaults(),
+			garageLayoutMgr:   glm.NewGarageLayoutManager(stateDir, ip),
 			currBootstrap:     currBootstrap,
 			workerCtx:         ctx,
 			workerCancel:      cancel,
@@ -237,6 +245,12 @@ func newNetwork(
 		return nil, fmt.Errorf("resolving network config: %w", err)
 	}
 
+	if err := n.garageLayoutMgr.Validate(
+		ctx, n.networkConfig.Storage.Allocations,
+	); err != nil {
+		return nil, ErrInvalidConfig.WithData(err.Error())
+	}
+
 	secretsDir, err := n.stateDir.MkChildDir("secrets", dirsMayExist)
 	if err != nil {
 		return nil, fmt.Errorf("creating secrets dir: %w", err)
@@ -399,6 +413,11 @@ func (constructorsImpl) create(
 		)
 	}
 
+	err = n.garageLayoutMgr.Validate(ctx, n.networkConfig.Storage.Allocations)
+	if err != nil {
+		return nil, ErrInvalidConfig.WithData(err.Error())
+	}
+
 	err = daecommon.SetGarageRPCSecret(ctx, n.secretsStore, garageRPCSecret)
 	if err != nil {
 		return nil, fmt.Errorf("setting garage RPC secret: %w", err)
@@ -411,6 +430,12 @@ func (constructorsImpl) create(
 		return nil, fmt.Errorf("setting nebula CA signing key secret: %w", err)
 	}
 
+	if err = n.garageLayoutMgr.SetActiveAllocations(
+		ctx, n.networkConfig.Storage.Allocations,
+	); err != nil {
+		return nil, fmt.Errorf("initializing GLM active allocations: %w", err)
+	}
+
 	if err := n.initialize(ctx, true); err != nil {
 		return nil, fmt.Errorf("initializing with bootstrap: %w", err)
 	}
@@ -418,10 +443,12 @@ func (constructorsImpl) create(
 	return n, nil
 }
 
-// preChildrenInit performs steps which are required prior to children being
-// initializes/reloaded. The lock must be held when this is called (if not being
-// called as part of initialize).
-func (n *network) preChildrenInit(ctx context.Context) error {
+// updateBootstrapUnsafe updates both the locally saved bootstrap as well as
+// this host's bootstrap host info in garage, first applying the network config
+// to the bootstrap host info.
+//
+// Must be called with the lock held.
+func (n *network) updateBootstrapUnsafe(ctx context.Context) error {
 	var err error
 	if n.currBootstrap, err = applyNetworkConfigToBootstrap(
 		n.networkConfig, n.currBootstrap,
@@ -436,61 +463,22 @@ func (n *network) preChildrenInit(ctx context.Context) error {
 		return fmt.Errorf("writing bootstrap to state dir: %w", err)
 	}
 
+	n.logger.Info(ctx, "Updating host info in garage")
+	if err := putGarageBoostrapHost(
+		ctx, n.secretsStore, n.currBootstrap,
+	); err != nil {
+		return fmt.Errorf("updating host info in garage: %w", err)
+	}
+
 	return nil
 }
 
-// postChildrenInit performs steps which are required after children have been
-// initialized/reloaded. The lock must be held when this is called (if not being
-// called as part of initialize).
-func (n *network) postChildrenInit(
-	ctx context.Context,
-	prevThisHost bootstrap.Host,
-	createGarageGlobalBucket bool,
+// see comment on garageWaitForAlloc
+func (n *network) garageWaitForAllocs(
+	ctx context.Context, allocs []daecommon.ConfigStorageAllocation,
 ) error {
-	var (
-		thisHost                     = n.currBootstrap.ThisHost()
-		garageAdminClient, hasGarage = n.children.GarageAdminClient()
-	)
-
-	if hasGarage {
-		defer garageAdminClient.Close()
-	}
-
-	if hasGarage {
-		n.logger.Info(ctx, "Applying garage layout")
-		if err := garageApplyLayout(
-			ctx,
-			n.logger,
-			n.networkConfig,
-			garageAdminClient,
-			prevThisHost, thisHost,
-		); err != nil {
-			return fmt.Errorf("applying garage layout: %w", err)
-		}
-	}
-
-	if createGarageGlobalBucket {
-		n.logger.Info(ctx, "Initializing garage shared global bucket")
-		garageGlobalBucketCreds, err := garageInitializeGlobalBucket(
-			ctx,
-			n.logger,
-			n.networkConfig,
-			garageAdminClient,
-			thisHost,
-		)
-		if err != nil {
-			return fmt.Errorf("initializing global bucket: %w", err)
-		}
-
-		err = daecommon.SetGarageS3APIGlobalBucketCredentials(
-			ctx, n.secretsStore, garageGlobalBucketCreds,
-		)
-		if err != nil {
-			return fmt.Errorf("storing global bucket creds: %w", err)
-		}
-	}
-
-	for _, alloc := range n.networkConfig.Storage.Allocations {
+	var errs []error
+	for _, alloc := range allocs {
 		garageAdminClient, ok := n.children.GarageAdminClientForAlloc(alloc)
 		if !ok {
 			return fmt.Errorf("no garage instance created for %+v", alloc)
@@ -502,21 +490,99 @@ func (n *network) postChildrenInit(
 		if err := garageWaitForAlloc(
 			ctx, n.logger, garageAdminClient,
 		); err != nil {
-			return fmt.Errorf(
+			errs = append(errs, fmt.Errorf(
 				"waiting for alloc %+v to initialize: %w", alloc, err,
-			)
+			))
 		}
 	}
 
-	n.logger.Info(ctx, "Updating host info in garage")
-	err := putGarageBoostrapHost(ctx, n.secretsStore, n.currBootstrap)
+	return errors.Join(errs...)
+}
+
+// must hold lock to call this
+func (n *network) glmStateTransitionUnsafe(ctx context.Context) error {
+	var knownNodes []garage.KnownNode
+
+	if adminClient, ok := n.children.GarageAdminClient(); ok {
+		defer adminClient.Close()
+
+		n.logger.Info(ctx, "Getting garage cluster status")
+		status, err := adminClient.Status(ctx)
+		if err != nil {
+			return fmt.Errorf("getting garage cluster state: %w", err)
+		}
+
+		knownNodes = status.Nodes
+	}
+
+	n.logger.Info(ctx, "Calculating garage layout state transition")
+	stateTx, err := n.garageLayoutMgr.CalculateStateTransition(
+		ctx, knownNodes, n.networkConfig.Storage.Allocations,
+	)
 	if err != nil {
-		return fmt.Errorf("updating host info in garage: %w", err)
+		return fmt.Errorf("getting next state tx: %w", err)
+	}
+
+	childrenNetworkConfig := n.networkConfig
+	childrenNetworkConfig.Storage.Allocations = stateTx.ActiveAllocations()
+
+	n.logger.Info(ctx, "Reloading children with updated storage allocations")
+	err = n.children.Reload(ctx, childrenNetworkConfig, n.currBootstrap)
+	if err != nil {
+		return fmt.Errorf("reloading children: %w", err)
+	}
+
+	if adminClient, ok := n.children.GarageAdminClient(); ok {
+		defer adminClient.Close()
+
+		var (
+			host = n.currBootstrap.ThisHost()
+			// From garage's perspective a node is "removed" from the cluster by
+			// removing its role, which puts it into the draining state.
+			removeIDs = stateTx.DrainAllocationIDs()
+		)
+
+		addModifyRoles, err := garageAllocsToRoles(
+			host, stateTx.AddModifyAllocations,
+		)
+		if err != nil {
+			return fmt.Errorf("converting allocs to roles: %w", err)
+		}
+
+		n.logger.Info(ctx, "Applying state transition to garage layout")
+		if err := adminClient.ApplyLayout(
+			ctx, addModifyRoles, removeIDs,
+		); err != nil {
+			return fmt.Errorf("applying state tx to layout: %w", err)
+		}
+	} else {
+		n.logger.Info(ctx, "No garage instances running, no layout changes to make")
+	}
+
+	if err := n.garageWaitForAllocs(
+		ctx, n.networkConfig.Storage.Allocations,
+	); err != nil {
+		return fmt.Errorf(
+			"waiting for garage allocations to fully initialize: %w", err,
+		)
+	}
+
+	n.logger.Info(ctx, "Committing state transition")
+	if err = n.garageLayoutMgr.CommitStateTransition(
+		ctx, stateTx,
+	); err != nil {
+		return fmt.Errorf("commiting state tx: %w", err)
 	}
 
 	return nil
 }
 
+func (n *network) glmStateTransition(ctx context.Context) error {
+	n.l.Lock()
+	defer n.l.Unlock()
+	return n.glmStateTransitionUnsafe(ctx)
+}
+
 func (n *network) reloadHosts(ctx context.Context) error {
 	n.l.RLock()
 	currBootstrap := n.currBootstrap
@@ -541,8 +607,13 @@ func (n *network) reloadHosts(ctx context.Context) error {
 		return fmt.Errorf("writing bootstrap to state dir: %w", err)
 	}
 
+	childrenNetworkConfig, err := n.getChildrenNetworkConfig(ctx)
+	if err != nil {
+		return fmt.Errorf("getting network config for children: %w", err)
+	}
+
 	n.logger.Info(ctx, "Reloading child processes")
-	err = n.children.Reload(ctx, n.networkConfig, n.currBootstrap)
+	err = n.children.Reload(ctx, childrenNetworkConfig, n.currBootstrap)
 	if err != nil {
 		return fmt.Errorf(
 			"reloading child processes: %w", err,
@@ -566,8 +637,8 @@ func (n *network) periodically(
 		ticker := time.NewTicker(period)
 		defer ticker.Stop()
 
-		n.logger.Info(ctx, "Starting background job runner")
-		defer n.logger.Info(ctx, "Stopping background job runner")
+		n.logger.Info(ctx, "Starting background job worker")
+		defer n.logger.Info(ctx, "Stopping background job worker")
 
 		for {
 			select {
@@ -575,7 +646,7 @@ func (n *network) periodically(
 				return
 
 			case <-ticker.C:
-				n.logger.Info(ctx, "Background job running")
+				n.logger.Info(ctx, "Background job worker")
 				if err := fn(ctx); err != nil {
 					n.logger.Error(ctx, "Background job failed", err)
 				}
@@ -584,14 +655,102 @@ func (n *network) periodically(
 	}()
 }
 
-func (n *network) initialize(ctx context.Context, isCreate bool) error {
-	var (
-		prevThisHost = n.currBootstrap.ThisHost()
-		err          error
-	)
+func (n *network) getChildrenNetworkConfig(
+	ctx context.Context,
+) (
+	daecommon.NetworkConfig, error,
+) {
+	childrenNetworkConfig := n.networkConfig
 
-	if err := n.preChildrenInit(ctx); err != nil {
-		return fmt.Errorf("performing pre-initialization: %w", err)
+	activeStorageAllocs, err := n.garageLayoutMgr.GetActiveAllocations(ctx)
+	if err != nil {
+		return daecommon.NetworkConfig{}, fmt.Errorf(
+			"getting active storage allocations: %w", err,
+		)
+	}
+	childrenNetworkConfig.Storage.Allocations = activeStorageAllocs
+
+	return childrenNetworkConfig, nil
+}
+
+func (n *network) initializePostChildren(
+	ctx context.Context, isCreate bool,
+) error {
+	if !isCreate {
+		n.logger.Info(ctx, "Making any necessary changes to garage layout")
+		if err := n.glmStateTransitionUnsafe(ctx); err != nil {
+			return fmt.Errorf("performing garage layout transition: %w", err)
+		}
+	} else {
+		roles, err := garageAllocsToRoles(
+			n.currBootstrap.ThisHost(), n.networkConfig.Storage.Allocations,
+		)
+		if err != nil {
+			return fmt.Errorf("converting allocs to roles: %w", err)
+		}
+
+		garageAdminClient, _ := n.children.GarageAdminClient()
+		defer garageAdminClient.Close()
+
+		n.logger.Info(ctx, "Applying initial garage layout")
+		if err := garageAdminClient.ApplyLayout(ctx, roles, nil); err != nil {
+			return fmt.Errorf("applying initial garage layout: %w", err)
+		}
+
+		n.logger.Info(ctx, "Initializing garage shared global bucket")
+		garageGlobalBucketCreds, err := garageInitializeGlobalBucket(
+			ctx, garageAdminClient,
+		)
+		if err != nil {
+			return fmt.Errorf("initializing global bucket: %w", err)
+		}
+
+		if err = daecommon.SetGarageS3APIGlobalBucketCredentials(
+			ctx, n.secretsStore, garageGlobalBucketCreds,
+		); err != nil {
+			return fmt.Errorf("storing global bucket creds: %w", err)
+		}
+
+		n.logger.Info(ctx, "Waiting for garage instances to finish initializing")
+		if err := n.garageWaitForAllocs(
+			ctx, n.networkConfig.Storage.Allocations,
+		); err != nil {
+			return fmt.Errorf(
+				"waiting for garage allocations to fully initialize: %w", err,
+			)
+		}
+	}
+
+	if err := n.updateBootstrapUnsafe(ctx); err != nil {
+		return fmt.Errorf("updating bootstrap: %w", err)
+	}
+
+	// Do this now so that everything is stable before returning. This also
+	// serves a dual-purpose, as it makes sure that the above bootstrap update
+	// has propagated from the local garage instance, if any.
+	n.logger.Info(ctx, "Reloading hosts from network storage")
+	if err := n.reloadHosts(ctx); err != nil {
+		return fmt.Errorf("Reloading network bootstrap: %w", err)
+	}
+
+	return nil
+}
+
+func (n *network) initialize(ctx context.Context, isCreate bool) error {
+	childrenNetworkConfig, err := n.getChildrenNetworkConfig(ctx)
+	if err != nil {
+		return fmt.Errorf("getting network config for children: %w", err)
+	}
+
+	var garageBootstrapPeers []garage.RemoteNode
+	if isCreate {
+		garageBootstrapPeers, err = garageInitAllocs(
+			n.currBootstrap.ThisHost().IP(),
+			n.networkConfig.Storage.Allocations,
+		)
+		if err != nil {
+			return fmt.Errorf("initializing storage allocations: %w", err)
+		}
 	}
 
 	n.logger.Info(ctx, "Creating child processes")
@@ -600,37 +759,29 @@ func (n *network) initialize(ctx context.Context, isCreate bool) error {
 		n.logger.WithNamespace("children"),
 		n.envBinDirPath,
 		n.secretsStore,
-		n.networkConfig,
+		childrenNetworkConfig,
 		n.runtimeDir,
 		n.opts.GarageAdminToken,
 		n.nebulaDeviceNamer,
 		n.currBootstrap,
 		&children.Opts{
-			GarageNewCluster: isCreate,
+			GarageNewCluster:      isCreate,
+			GarageBootstrapPeers:  garageBootstrapPeers,
+			GarageDefaultDBEngine: n.opts.garageDefaultDBEngine,
 		},
 	)
 	if err != nil {
 		return fmt.Errorf("creating child processes: %w", err)
 	}
 
-	createGarageGlobalBucket := isCreate
-	err = n.postChildrenInit(ctx, prevThisHost, createGarageGlobalBucket)
-	if err != nil {
-		n.logger.Error(ctx, "Post-initialization failed, stopping child processes", err)
+	if err := n.initializePostChildren(ctx, isCreate); err != nil {
+		n.logger.Error(ctx, "Failed to initialize Network, shutting down children", err)
 		n.children.Shutdown()
-		return fmt.Errorf("performing post-initialization: %w", err)
-	}
-
-	// Do this now so that everything is stable before returning. This also
-	// serves a dual-purpose, as it makes sure that the PUT from the
-	// postChildrenInit above has propagated from the local garage instance, if
-	// there is one.
-	n.logger.Info(ctx, "Reloading hosts from network storage")
-	if err = n.reloadHosts(ctx); err != nil {
-		return fmt.Errorf("Reloading network bootstrap: %w", err)
+		return err
 	}
 
 	n.periodically("reloadHosts", n.reloadHosts, 3*time.Minute)
+	n.periodically("glmStateTransition", n.glmStateTransition, 10*time.Minute)
 
 	return nil
 }
@@ -924,10 +1075,11 @@ func (n *network) SetConfig(
 	n.l.Lock()
 	defer n.l.Unlock()
 
-	var (
-		prevThisHost = n.currBootstrap.ThisHost()
-		err          error
-	)
+	if err := n.garageLayoutMgr.Validate(
+		ctx, config.Storage.Allocations,
+	); err != nil {
+		return ErrInvalidConfig.WithData(err.Error())
+	}
 
 	if _, err := loadStoreConfig(n.stateDir, &config); err != nil {
 		return fmt.Errorf("storing new config: %w", err)
@@ -935,18 +1087,13 @@ func (n *network) SetConfig(
 
 	n.networkConfig = config
 
-	if err := n.preChildrenInit(ctx); err != nil {
-		return fmt.Errorf("performing pre-initialization: %w", err)
+	n.logger.Info(ctx, "Making any necessary changes to garage layout")
+	if err := n.glmStateTransitionUnsafe(ctx); err != nil {
+		return fmt.Errorf("performing garage layout transition: %w", err)
 	}
 
-	n.logger.Info(ctx, "Reloading child processes")
-	err = n.children.Reload(ctx, n.networkConfig, n.currBootstrap)
-	if err != nil {
-		return fmt.Errorf("reloading child processes: %w", err)
-	}
-
-	if err := n.postChildrenInit(ctx, prevThisHost, false); err != nil {
-		return fmt.Errorf("performing post-initialization: %w", err)
+	if err := n.updateBootstrapUnsafe(ctx); err != nil {
+		return fmt.Errorf("updating bootstrap: %w", err)
 	}
 
 	return nil
diff --git a/go/daemon/network/network_it_test.go b/go/daemon/network/network_it_test.go
index 9d0bc0d..a43c2a9 100644
--- a/go/daemon/network/network_it_test.go
+++ b/go/daemon/network/network_it_test.go
@@ -15,7 +15,6 @@ import (
 	"time"
 
 	"github.com/stretchr/testify/assert"
-	"github.com/stretchr/testify/require"
 )
 
 func TestCreate(t *testing.T) {
@@ -59,6 +58,42 @@ func TestLoad(t *testing.T) {
 
 		assert.Equal(t, networkConfig, network.getConfig(t))
 	})
+
+	t.Run("garage lmdb db engine", func(t *testing.T) {
+		var (
+			h       = newIntegrationHarness(t)
+			network = h.createNetwork(t, "primus", &createNetworkOpts{
+				garageDefaultDBEngine: garagesrv.DBEngineLMDB,
+			})
+			metaPath = h.mkDir(t, "meta").Path
+		)
+
+		h.logger.Info(h.ctx, "Checking that garage is using the expected db engine")
+		garageConfig, err := os.ReadFile(
+			filepath.Join(network.runtimeDir.Path, "garage-3900.toml"),
+		)
+		assert.NoError(t, err)
+		assert.Contains(t,
+			string(garageConfig),
+			`db_engine = "`+garagesrv.DBEngineLMDB+`"`,
+		)
+		assert.NoFileExists(t, filepath.Join(metaPath, "db.sqlite"))
+
+		network.opts.garageDefaultDBEngine = ""
+		network.restart(t)
+
+		h.logger.Info(h.ctx, "Checking that garage is still using the expected db engine")
+		garageConfig, err = os.ReadFile(
+			filepath.Join(network.runtimeDir.Path, "garage-3900.toml"),
+		)
+		assert.NoError(t, err)
+		assert.Contains(t,
+			string(garageConfig),
+			`db_engine = "`+garagesrv.DBEngineLMDB+`"`,
+		)
+		assert.NoFileExists(t, filepath.Join(metaPath, "db.sqlite"))
+	})
+
 }
 
 func TestJoin(t *testing.T) {
@@ -85,7 +120,7 @@ func TestJoin(t *testing.T) {
 			})
 		)
 
-		t.Log("reloading primus' hosts")
+		h.logger.Info(h.ctx, "reloading primus' hosts")
 		assert.NoError(t, primus.Network.(*network).reloadHosts(h.ctx))
 
 		assert.Equal(t, primus.getHostsByName(t), secondus.getHostsByName(t))
@@ -94,57 +129,6 @@ func TestJoin(t *testing.T) {
 			secondus: 1,
 		})
 	})
-
-	// Assert that if primus runs the orphan remover at the same moment that
-	// secondus is joining that the layout applied by secondus doesn't get
-	// overwritten.
-	t.Run("with alloc/remove orphans after garage layout applied", func(t *testing.T) {
-		t.Skip("This is currently expected to fail. Orphan removal is going to be reworked accordingly")
-
-		var (
-			h                 = newIntegrationHarness(t)
-			primus            = h.createNetwork(t, "primus", nil)
-			primusAdminClient = primus.garageAdminClient(t)
-			secondusBlocker   = toolkit.NewTestBlocker(t)
-		)
-
-		secondusBlocker.ExpectBlockpoint("garageLayoutApplied").On(
-			t, h.ctx, func() {
-				h.logger.Info(h.ctx, "Waiting for new layout to propagate to primus")
-				err := toolkit.UntilTrue(
-					h.ctx, h.logger, 1*time.Second, func() (bool, error) {
-						layout, err := primusAdminClient.GetLayout(h.ctx)
-						if err != nil {
-							return false, fmt.Errorf("getting layout: %w", err)
-						}
-
-						return len(layout.Roles) == 4, nil
-					},
-				)
-
-				if !assert.NoError(t, err) {
-					return
-				}
-
-				//h.logger.Info(h.ctx, "Calling removeOrphanGarageNodes")
-				//assert.NoError(
-				//	t, primus.Network.(*network).removeOrphanGarageNodes(h.ctx),
-				//)
-			},
-		)
-
-		secondus := h.joinNetwork(t, primus, "secondus", &joinNetworkOpts{
-			networkConfigOpts: &networkConfigOpts{
-				numStorageAllocs: 1,
-			},
-			blocker: secondusBlocker,
-		})
-
-		assertGarageLayout(t, map[*integrationHarnessNetwork]int{
-			primus:   3,
-			secondus: 1,
-		})
-	})
 }
 
 func TestNetwork_GetBootstrap(t *testing.T) {
@@ -220,7 +204,7 @@ func TestNetwork_SetConfig(t *testing.T) {
 
 		assert.NoError(t, network.SetConfig(h.ctx, networkConfig))
 
-		t.Log("Checking that the Host information was updated")
+		h.logger.Info(h.ctx, "Checking that the Host information was updated")
 		newHostsByName := network.getHostsByName(t)
 		newHost, ok := newHostsByName[network.hostName]
 		assert.True(t, ok)
@@ -236,20 +220,20 @@ func TestNetwork_SetConfig(t *testing.T) {
 			RPCPort:   4901,
 		}, newAlloc)
 
-		t.Log("Checking that the bootstrap file was written with the new host config")
+		h.logger.Info(h.ctx, "Checking that the bootstrap file was written with the new host config")
 		var storedBootstrap bootstrap.Bootstrap
 		assert.NoError(t, jsonutil.LoadFile(
 			&storedBootstrap, bootstrap.StateDirPath(network.stateDir.Path),
 		))
 		assert.Equal(t, newHostsByName, storedBootstrap.Hosts)
 
-		t.Log("Checking that garage layout contains the new allocation")
+		h.logger.Info(h.ctx, "Checking that garage layout contains the new allocation")
 		expRoles := allocsToRoles(network.hostName, allocs)
 		layout, err := network.garageAdminClient(t).GetLayout(h.ctx)
 		assert.NoError(t, err)
 		assert.ElementsMatch(t, expRoles, layout.Roles)
 
-		t.Log("Checking that garage is using the expected db engine")
+		h.logger.Info(h.ctx, "Checking that garage is using the expected db engine")
 		garageConfig, err := os.ReadFile(
 			filepath.Join(network.runtimeDir.Path, "garage-4901.toml"),
 		)
@@ -261,46 +245,6 @@ func TestNetwork_SetConfig(t *testing.T) {
 		assert.FileExists(t, filepath.Join(metaPath, "db.sqlite"))
 	})
 
-	t.Run("add storage alloc/lmdb", func(t *testing.T) {
-		var (
-			h             = newIntegrationHarness(t)
-			network       = h.createNetwork(t, "primus", nil)
-			networkConfig = network.getConfig(t)
-			dataPath      = h.mkDir(t, "data").Path
-			metaPath      = h.mkDir(t, "meta").Path
-		)
-
-		networkConfig.Storage.Allocations = append(
-			networkConfig.Storage.Allocations,
-			daecommon.ConfigStorageAllocation{
-				DataPath:  dataPath,
-				MetaPath:  metaPath,
-				Capacity:  1,
-				S3APIPort: 4900,
-				RPCPort:   4901,
-				AdminPort: 4902,
-			},
-		)
-
-		// Creating the directory is enough to ensure that Isle chooses LMDB as
-		// the db engine.
-		lmdbPath := filepath.Join(metaPath, "db.lmdb")
-		require.NoError(t, os.Mkdir(lmdbPath, 0755))
-
-		assert.NoError(t, network.SetConfig(h.ctx, networkConfig))
-
-		t.Log("Checking that garage is using the expected db engine")
-		garageConfig, err := os.ReadFile(
-			filepath.Join(network.runtimeDir.Path, "garage-4901.toml"),
-		)
-		assert.NoError(t, err)
-		assert.Contains(t,
-			string(garageConfig),
-			`db_engine = "`+garagesrv.DBEngineLMDB+`"`,
-		)
-		assert.NoFileExists(t, filepath.Join(metaPath, "db.sqlite"))
-	})
-
 	t.Run("remove storage alloc", func(t *testing.T) {
 		var (
 			h       = newIntegrationHarness(t)
@@ -309,17 +253,21 @@ func TestNetwork_SetConfig(t *testing.T) {
 			})
 			networkConfig = network.getConfig(t)
 
-			prevHost          = network.getHostsByName(t)[network.hostName]
-			removedAlloc      = networkConfig.Storage.Allocations[3]
-			removedGarageInst = daecommon.BootstrapGarageHostForAlloc(
-				prevHost, removedAlloc,
-			)
+			prevHost     = network.getHostsByName(t)[network.hostName]
+			removedAlloc = networkConfig.Storage.Allocations[3]
 		)
 
+		var removedGarageInst bootstrap.GarageHostInstance
+		for _, removedGarageInst = range prevHost.Garage.Instances {
+			if removedGarageInst.RPCPort == removedAlloc.RPCPort {
+				break
+			}
+		}
+
 		networkConfig.Storage.Allocations = networkConfig.Storage.Allocations[:3]
 		assert.NoError(t, network.SetConfig(h.ctx, networkConfig))
 
-		t.Log("Checking that the Host information was updated")
+		h.logger.Info(h.ctx, "Checking that the Host information was updated")
 		newHostsByName := network.getHostsByName(t)
 		newHost, ok := newHostsByName[network.hostName]
 		assert.True(t, ok)
@@ -328,68 +276,20 @@ func TestNetwork_SetConfig(t *testing.T) {
 		assert.Len(t, allocs, 3)
 		assert.NotContains(t, allocs, removedGarageInst)
 
-		t.Log("Checking that the bootstrap file was written with the new host config")
+		h.logger.Info(h.ctx, "Checking that the bootstrap file was written with the new host config")
 		var storedBootstrap bootstrap.Bootstrap
 		assert.NoError(t, jsonutil.LoadFile(
 			&storedBootstrap, bootstrap.StateDirPath(network.stateDir.Path),
 		))
 		assert.Equal(t, newHostsByName, storedBootstrap.Hosts)
 
-		t.Log("Checking that garage layout contains the new allocation")
+		h.logger.Info(h.ctx, "Checking that garage layout contains the new allocation")
 		expRoles := allocsToRoles(network.hostName, allocs)
 		layout, err := network.garageAdminClient(t).GetLayout(h.ctx)
 		assert.NoError(t, err)
 		assert.ElementsMatch(t, expRoles, layout.Roles)
 	})
 
-	t.Run("remove all storage allocs", func(t *testing.T) {
-		t.Skip("This is currently expected to fail. Orphan removal is going to be reworked accordingly")
-
-		var (
-			h        = newIntegrationHarness(t)
-			primus   = h.createNetwork(t, "primus", nil)
-			secondus = h.joinNetwork(t, primus, "secondus", &joinNetworkOpts{
-				networkConfigOpts: &networkConfigOpts{
-					numStorageAllocs: 1,
-				},
-			})
-			networkConfig = secondus.getConfig(t)
-
-			prevHost    = secondus.getHostsByName(t)[secondus.hostName]
-			removedRole = allocsToRoles(
-				secondus.hostName, prevHost.Garage.Instances,
-			)[0]
-
-			primusGarageAdminClient = primus.garageAdminClient(t)
-		)
-
-		networkConfig.Storage.Allocations = nil
-		assert.NoError(t, secondus.SetConfig(h.ctx, networkConfig))
-
-		t.Log("Checking that the Host information was updated")
-		newHostsByName := primus.getHostsByName(t)
-		newHost, ok := newHostsByName[secondus.hostName]
-		assert.True(t, ok)
-
-		allocs := newHost.HostConfigured.Garage.Instances
-		assert.Empty(t, allocs)
-
-		t.Log("Checking that garage layout still contains the old allocation")
-		layout, err := primusGarageAdminClient.GetLayout(h.ctx)
-		assert.NoError(t, err)
-		assert.Contains(t, layout.Roles, removedRole)
-
-		//t.Log("Removing orphan garage nodes with primus")
-		//assert.NoError(
-		//	t, primus.Network.(*network).removeOrphanGarageNodes(h.ctx),
-		//)
-
-		t.Log("Checking that garage layout no longer contains the old allocation")
-		layout, err = primusGarageAdminClient.GetLayout(h.ctx)
-		assert.NoError(t, err)
-		assert.NotContains(t, layout.Roles, removedRole)
-	})
-
 	t.Run("changes reflected after restart", func(t *testing.T) {
 		var (
 			h       = newIntegrationHarness(t)
@@ -408,3 +308,56 @@ func TestNetwork_SetConfig(t *testing.T) {
 		assert.Equal(t, networkConfig, network.getConfig(t))
 	})
 }
+
+func TestNetwork_glmStateTransition(t *testing.T) {
+	var (
+		h        = newIntegrationHarness(t)
+		primus   = h.createNetwork(t, "primus", nil)
+		secondus = h.joinNetwork(t, primus, "secondus", &joinNetworkOpts{
+			networkConfigOpts: &networkConfigOpts{
+				numStorageAllocs: 1,
+			},
+		})
+		secondusNetworkConfig = secondus.getConfig(t)
+		secondusAdminClient   = secondus.garageAdminClient(t)
+		secondusNetworkDirect = secondus.Network.(*network)
+		secondsBootstrapHost  = primus.getHostsByName(t)["secondus"]
+	)
+
+	assertGarageLayout(t, map[*integrationHarnessNetwork]int{
+		primus:   3,
+		secondus: 1,
+	})
+
+	secondusNetworkConfig.Storage.Allocations = nil
+	assert.NoError(t, secondus.SetConfig(h.ctx, secondusNetworkConfig))
+
+	assert.Len(t, secondusNetworkDirect.children.ActiveStorageAllocations(), 1)
+
+	h.logger.Info(h.ctx, "Waiting for secondus to finish draining")
+	err := toolkit.UntilTrue(
+		h.ctx, h.logger, 1*time.Second, func() (bool, error) {
+			status, err := secondusAdminClient.Status(h.ctx)
+			if err != nil {
+				return false, fmt.Errorf("getting status: %w", err)
+			}
+
+			for _, node := range status.Nodes {
+				if node.Addr.Addr() == secondsBootstrapHost.IP() {
+					return !node.Draining, nil
+				}
+			}
+
+			return false, fmt.Errorf("secondus not found in cluster status: %+v", status)
+		},
+	)
+	assert.NoError(t, err)
+
+	h.logger.Info(h.ctx, "Running GLM state transition")
+	assert.NoError(t, secondusNetworkDirect.glmStateTransition(h.ctx))
+
+	assertGarageLayout(t, map[*integrationHarnessNetwork]int{
+		primus: 3,
+	})
+	assert.Empty(t, secondusNetworkDirect.children.ActiveStorageAllocations())
+}
diff --git a/go/daemon/network/network_it_util_test.go b/go/daemon/network/network_it_util_test.go
index 39a7961..62b4789 100644
--- a/go/daemon/network/network_it_util_test.go
+++ b/go/daemon/network/network_it_util_test.go
@@ -8,6 +8,7 @@ import (
 	"isle/daemon/children"
 	"isle/daemon/daecommon"
 	"isle/garage"
+	"isle/garage/garagesrv"
 	"isle/nebula"
 	"isle/toolkit"
 	"os"
@@ -17,6 +18,7 @@ import (
 	"sync/atomic"
 	"testing"
 
+	"dev.mediocregopher.com/mediocre-go-lib.git/mctx"
 	"dev.mediocregopher.com/mediocre-go-lib.git/mlog"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -75,24 +77,32 @@ func newIntegrationHarness(t *testing.T) *integrationHarness {
 	t.Parallel()
 	toolkit.MarkIntegrationTest(t)
 
+	var (
+		ctx    = context.Background()
+		logger = toolkit.NewTestLogger(t)
+	)
+
 	rootDir, err := os.MkdirTemp("", "isle-network-it-test.*")
 	require.NoError(t, err)
 
-	t.Logf("Temporary test directory: %q", rootDir)
+	{
+		ctx := mctx.Annotate(ctx, "rootDir", rootDir)
+		logger.Info(ctx, "Created temporary test root directory")
 
-	t.Cleanup(func() {
-		if t.Failed() {
-			t.Logf("Temp directory for failed test not deleted: %q", rootDir)
-			return
-		}
+		t.Cleanup(func() {
+			if t.Failed() {
+				logger.Info(ctx, "Test failed, temporarty test root directory NOT deleted")
+				return
+			}
 
-		t.Logf("Deleting temp directory %q", rootDir)
-		assert.NoError(t, os.RemoveAll(rootDir))
-	})
+			logger.Info(ctx, "Deleting temporary test root directory")
+			assert.NoError(t, os.RemoveAll(rootDir))
+		})
+	}
 
 	return &integrationHarness{
-		ctx:               context.Background(),
-		logger:            toolkit.NewTestLogger(t),
+		ctx:               ctx,
+		logger:            logger,
 		constructors:      newConstructors(),
 		rootDir:           toolkit.Dir{Path: rootDir},
 		nebulaDeviceNamer: children.NewNebulaDeviceNamer(),
@@ -102,7 +112,10 @@ func newIntegrationHarness(t *testing.T) *integrationHarness {
 func (h *integrationHarness) mkDir(t *testing.T, name string) toolkit.Dir {
 	fullName := fmt.Sprintf("%s-%d", name, h.dirCounter.Add(1)-1)
 
-	t.Logf("Creating directory %q", fullName)
+	h.logger.Info(
+		mctx.Annotate(h.ctx, "dirName", fullName),
+		"Creating directory",
+	)
 	d, err := h.rootDir.MkChildDir(fullName, false)
 	require.NoError(t, err)
 
@@ -149,9 +162,10 @@ func (h *integrationHarness) mkNetworkConfig(
 }
 
 type createNetworkOpts struct {
-	creationParams   bootstrap.CreationParams
-	manualShutdown   bool
-	numStorageAllocs int
+	creationParams        bootstrap.CreationParams
+	manualShutdown        bool
+	numStorageAllocs      int
+	garageDefaultDBEngine garagesrv.DBEngine
 }
 
 func (o *createNetworkOpts) withDefaults() *createNetworkOpts {
@@ -187,11 +201,10 @@ func (h *integrationHarness) createNetwork(
 	hostNameStr string,
 	opts *createNetworkOpts,
 ) *integrationHarnessNetwork {
-	t.Logf("Creating as %q", hostNameStr)
 	opts = opts.withDefaults()
 
 	var (
-		logger        = h.logger.WithNamespace("network").WithNamespace(hostNameStr)
+		logger        = h.logger.WithNamespace("networks").WithNamespace(hostNameStr)
 		networkConfig = h.mkNetworkConfig(t, &networkConfigOpts{
 			hasPublicAddr:    true,
 			numStorageAllocs: opts.numStorageAllocs,
@@ -204,11 +217,13 @@ func (h *integrationHarness) createNetwork(
 		hostName = nebula.HostName(hostNameStr)
 
 		networkOpts = &Opts{
-			GarageAdminToken: "admin_token",
-			Config:           &networkConfig,
+			GarageAdminToken:      "admin_token",
+			Config:                &networkConfig,
+			garageDefaultDBEngine: opts.garageDefaultDBEngine,
 		}
 	)
 
+	logger.Info(h.ctx, "Creating network")
 	network, err := h.constructors.create(
 		h.ctx,
 		logger,
@@ -239,9 +254,9 @@ func (h *integrationHarness) createNetwork(
 
 	if !opts.manualShutdown {
 		t.Cleanup(func() {
-			t.Logf("Shutting down Network %q", hostNameStr)
+			logger.Info(nh.ctx, "Shutting down Network")
 			if err := nh.Shutdown(); err != nil {
-				t.Logf("Shutting down Network %q failed: %v", hostNameStr, err)
+				logger.Error(nh.ctx, "Shutting down Network failed", err)
 			}
 		})
 	}
@@ -273,18 +288,24 @@ func (h *integrationHarness) joinNetwork(
 	opts *joinNetworkOpts,
 ) *integrationHarnessNetwork {
 	opts = opts.withDefaults()
-	hostName := nebula.HostName(hostNameStr)
 
-	t.Logf("Creating bootstrap for %q", hostNameStr)
-	joiningBootstrap, err := network.CreateHost(h.ctx, hostName, CreateHostOpts{
-		CanCreateHosts: opts.canCreateHosts,
-	})
+	var (
+		hostName      = nebula.HostName(hostNameStr)
+		createHostCtx = mctx.Annotate(h.ctx, "hostName", hostName)
+	)
+
+	h.logger.Info(createHostCtx, "Creating bootstrap")
+	joiningBootstrap, err := network.CreateHost(
+		createHostCtx, hostName, CreateHostOpts{
+			CanCreateHosts: opts.canCreateHosts,
+		},
+	)
 	if err != nil {
 		t.Fatalf("creating host joining bootstrap: %v", err)
 	}
 
 	var (
-		logger        = h.logger.WithNamespace("network").WithNamespace(hostNameStr)
+		logger        = h.logger.WithNamespace("networks").WithNamespace(hostNameStr)
 		networkConfig = h.mkNetworkConfig(t, opts.networkConfigOpts)
 		stateDir      = h.mkDir(t, "state")
 		runtimeDir    = h.mkDir(t, "runtime")
@@ -295,7 +316,7 @@ func (h *integrationHarness) joinNetwork(
 		}
 	)
 
-	t.Logf("Joining as %q", hostNameStr)
+	logger.Info(h.ctx, "Joining")
 	joinedNetwork, err := h.constructors.join(
 		h.ctx,
 		logger,
@@ -324,9 +345,9 @@ func (h *integrationHarness) joinNetwork(
 
 	if !opts.manualShutdown {
 		t.Cleanup(func() {
-			t.Logf("Shutting down Network %q", hostNameStr)
+			nh.logger.Info(nh.ctx, "Shutting down Network")
 			if err := nh.Shutdown(); err != nil {
-				t.Logf("Shutting down Network %q failed: %v", hostNameStr, err)
+				nh.logger.Error(nh.ctx, "Shutting down Network failed", err)
 			}
 		})
 	}
@@ -335,10 +356,10 @@ func (h *integrationHarness) joinNetwork(
 }
 
 func (nh *integrationHarnessNetwork) restart(t *testing.T) {
-	t.Log("Shutting down network (restart)")
+	nh.logger.Info(nh.ctx, "Shutting down network (restart)")
 	require.NoError(t, nh.Network.Shutdown())
 
-	t.Log("Loading network (restart)")
+	nh.logger.Info(nh.ctx, "Loading network (restart)")
 	var err error
 	nh.Network, err = nh.constructors.load(
 		nh.ctx,
diff --git a/go/garage/admin_client.go b/go/garage/admin_client.go
index 005c9f9..e63a546 100644
--- a/go/garage/admin_client.go
+++ b/go/garage/admin_client.go
@@ -141,6 +141,7 @@ type KnownNode struct {
 	Addr            netip.AddrPort `json:"addr"`
 	IsUp            bool           `json:"isUp"`
 	LastSeenSecsAgo int            `json:"lastSeenSecsAgo"`
+	Draining        bool           `json:"draining"`
 	HostName        string         `json:"hostname"`
 }
 
diff --git a/go/garage/garagesrv/garagesrv.go b/go/garage/garagesrv/garagesrv.go
index be9a435..3df7d4c 100644
--- a/go/garage/garagesrv/garagesrv.go
+++ b/go/garage/garagesrv/garagesrv.go
@@ -11,7 +11,6 @@ import (
 	"io/fs"
 	"os"
 	"path/filepath"
-	"strconv"
 )
 
 // DBEngine enumerates the garage db engines which are supported by Isle.
@@ -31,13 +30,9 @@ func nodeKeyPubPath(metaDirPath string) string {
 	return filepath.Join(metaDirPath, "node_key.pub")
 }
 
-func nodeRPCPortPath(metaDirPath string) string {
-	return filepath.Join(metaDirPath, "isle", "rpc_port")
-}
-
-// loadAllocID returns the peer ID (ie the public key) of the node at the given
+// LoadAllocID returns the peer ID (ie the public key) of the node at the given
 // meta directory.
-func loadAllocID(metaDirPath string) (string, error) {
+func LoadAllocID(metaDirPath string) (string, error) {
 	nodeKeyPubPath := nodeKeyPubPath(metaDirPath)
 
 	pubKey, err := os.ReadFile(nodeKeyPubPath)
@@ -61,8 +56,8 @@ func generatePeerKey() (pubKey, privKey []byte) {
 
 // InitAlloc initializes the meta directory and keys for a particular
 // allocation, if it hasn't been done so already. It returns the peer ID (ie the
-// public key) and the rpc port in any case.
-func InitAlloc(metaDirPath string, initRPCPort int) (string, int, error) {
+// public key) in any case.
+func InitAlloc(metaDirPath string) (string, error) {
 
 	initDirFor := func(path string) error {
 		dir := filepath.Dir(path)
@@ -70,110 +65,97 @@ func InitAlloc(metaDirPath string, initRPCPort int) (string, int, error) {
 	}
 
 	var err error
-
 	exists := func(path string) bool {
-
 		if err != nil {
 			return false
-
 		} else if _, err = os.Stat(path); errors.Is(err, fs.ErrNotExist) {
 			err = nil
 			return false
-
 		} else if err != nil {
 			err = fmt.Errorf("checking if %q exists: %w", path, err)
 			return false
 		}
-
 		return true
 	}
 
-	nodeKeyPath := nodeKeyPath(metaDirPath)
-	nodeKeyPubPath := nodeKeyPubPath(metaDirPath)
-	nodeRPCPortPath := nodeRPCPortPath(metaDirPath)
-
-	nodeKeyPathExists := exists(nodeKeyPath)
-	nodeKeyPubPathExists := exists(nodeKeyPubPath)
-	nodeRPCPortPathExists := exists(nodeRPCPortPath)
-
-	if err != nil {
-		return "", 0, err
-
-	} else if nodeKeyPubPathExists != nodeKeyPathExists {
-		return "", 0, fmt.Errorf("%q or %q exist without the other existing", nodeKeyPath, nodeKeyPubPath)
-
-	}
-
 	var (
-		pubKeyStr string
-		rpcPort   int
+		nodeKeyPath    = nodeKeyPath(metaDirPath)
+		nodeKeyPubPath = nodeKeyPubPath(metaDirPath)
+
+		nodeKeyPathExists    = exists(nodeKeyPath)
+		nodeKeyPubPathExists = exists(nodeKeyPubPath)
 	)
 
-	if nodeKeyPathExists {
+	if err != nil {
+		return "", err
+	} else if nodeKeyPubPathExists != nodeKeyPathExists {
+		return "", fmt.Errorf(
+			"%q or %q exist without the other existing",
+			nodeKeyPath,
+			nodeKeyPubPath,
+		)
+	}
 
-		if pubKeyStr, err = loadAllocID(metaDirPath); err != nil {
-			return "", 0, fmt.Errorf("reading node public key file: %w", err)
+	var pubKeyStr string
+	if nodeKeyPathExists {
+		if pubKeyStr, err = LoadAllocID(metaDirPath); err != nil {
+			return "", fmt.Errorf("reading node public key file: %w", err)
 		}
 
 	} else {
-
 		if err := initDirFor(nodeKeyPath); err != nil {
-			return "", 0, fmt.Errorf("creating directory for %q: %w", nodeKeyPath, err)
+			return "", fmt.Errorf(
+				"creating directory for %q: %w", nodeKeyPath, err,
+			)
 		}
 
 		pubKey, privKey := generatePeerKey()
-
 		if err := os.WriteFile(nodeKeyPath, privKey, 0400); err != nil {
-			return "", 0, fmt.Errorf("writing private key to %q: %w", nodeKeyPath, err)
+			return "", fmt.Errorf(
+				"writing private key to %q: %w", nodeKeyPath, err,
+			)
 
 		} else if err := os.WriteFile(nodeKeyPubPath, pubKey, 0440); err != nil {
-			return "", 0, fmt.Errorf("writing public key to %q: %w", nodeKeyPubPath, err)
+			return "", fmt.Errorf(
+				"writing public key to %q: %w", nodeKeyPubPath, err,
+			)
 		}
 
 		pubKeyStr = hex.EncodeToString(pubKey)
 	}
 
-	if nodeRPCPortPathExists {
-
-		if rpcPortStr, err := os.ReadFile(nodeRPCPortPath); err != nil {
-			return "", 0, fmt.Errorf("reading rpc port from %q: %w", nodeRPCPortPath, err)
-
-		} else if rpcPort, err = strconv.Atoi(string(rpcPortStr)); err != nil {
-			return "", 0, fmt.Errorf("parsing rpc port %q from %q: %w", rpcPortStr, nodeRPCPortPath, err)
-		}
-
-	} else {
-
-		if err := initDirFor(nodeRPCPortPath); err != nil {
-			return "", 0, fmt.Errorf("creating directory for %q: %w", nodeRPCPortPath, err)
-		}
-
-		rpcPortStr := strconv.Itoa(initRPCPort)
-
-		if err := os.WriteFile(nodeRPCPortPath, []byte(rpcPortStr), 0440); err != nil {
-			return "", 0, fmt.Errorf("writing rpc port %q to %q: %w", rpcPortStr, nodeRPCPortPath, err)
-		}
-
-		rpcPort = initRPCPort
-	}
-
-	return pubKeyStr, rpcPort, nil
+	return pubKeyStr, nil
 }
 
 // GetDBEngine returns the DBEngine currently being used at a particular meta
-// data directory. Defaults to DBEngineSqlite if the directory doesn't exist or
+// data directory, or returns the default if the directory doesn't exist or
 // hasn't been fully initialized yet.
-func GetDBEngine(metaDirPath string) (DBEngine, error) {
-	dbLMDBPath := filepath.Join(metaDirPath, "db.lmdb")
-
-	stat, err := os.Stat(dbLMDBPath)
-	if errors.Is(err, fs.ErrNotExist) {
-		return DBEngineSqlite, nil
-	} else if err != nil {
-		return "", fmt.Errorf("checking if %q exists: %w", dbLMDBPath, err)
-	} else if stat.IsDir() {
-		return DBEngineLMDB, nil
+func GetDBEngine(
+	metaDirPath string, defaultDBEngine DBEngine,
+) (
+	DBEngine, error,
+) {
+	search := []struct {
+		dbEngine  DBEngine
+		path      string
+		pathIsDir bool
+	}{
+		{DBEngineLMDB, filepath.Join(metaDirPath, "db.lmdb"), true},
+		{DBEngineSqlite, filepath.Join(metaDirPath, "db.sqlite"), false},
 	}
 
-	return DBEngineSqlite, nil
+	for _, s := range search {
+		stat, err := os.Stat(s.path)
+		if errors.Is(err, fs.ErrNotExist) {
+			continue
+		} else if err != nil {
+			return "", fmt.Errorf("checking if %q exists: %w", s.path, err)
+		} else if stat.IsDir() != s.pathIsDir {
+			continue
+		}
+
+		return s.dbEngine, nil
+	}
+
+	return defaultDBEngine, nil
 }
diff --git a/go/garage/peer.go b/go/garage/peer.go
index 4481281..a0a35f1 100644
--- a/go/garage/peer.go
+++ b/go/garage/peer.go
@@ -6,25 +6,28 @@ import (
 	"strconv"
 )
 
-// RemoteNode describes all information necessary to connect to a given garage
-// node.
-type RemoteNode struct {
-	ID        string
+// Node describes the common fields of both a [RemoteNode] and [LocalNode]
+type Node struct {
 	IP        string
 	RPCPort   int
 	S3APIPort int
 }
 
-// LocalNode describes the configuration of a local garage instance.
-type LocalNode struct {
-	RemoteNode
-
-	AdminPort int
+// RPCAddr returns the address of the node's RPC port.
+func (p Node) RPCAddr() string {
+	return net.JoinHostPort(p.IP, strconv.Itoa(p.RPCPort))
 }
 
-// RPCAddr returns the address of the node's RPC port.
-func (p RemoteNode) RPCAddr() string {
-	return net.JoinHostPort(p.IP, strconv.Itoa(p.RPCPort))
+// S3APIAddr returns the address of the node's S3 API port.
+func (p Node) S3APIAddr() string {
+	return net.JoinHostPort(p.IP, strconv.Itoa(p.S3APIPort))
+}
+
+// RemoteNode describes all information necessary to connect to a given garage
+// node.
+type RemoteNode struct {
+	Node
+	ID string
 }
 
 // RPCNodeAddr returns the full node address (e.g. "id@ip:port") of the garage
@@ -33,9 +36,10 @@ func (p RemoteNode) RPCNodeAddr() string {
 	return fmt.Sprintf("%s@%s", p.ID, p.RPCAddr())
 }
 
-// S3APIAddr returns the address of the node's S3 API port.
-func (p RemoteNode) S3APIAddr() string {
-	return net.JoinHostPort(p.IP, strconv.Itoa(p.S3APIPort))
+// LocalNode describes the configuration of a local garage instance.
+type LocalNode struct {
+	Node
+	AdminPort int
 }
 
 // AdminAddr returns the address of the node's S3 API port.
diff --git a/tasks/bugs/validate-non-conflicting-ports-or-dirs.md b/tasks/bugs/validate-non-conflicting-ports-or-dirs.md
new file mode 100644
index 0000000..95b27af
--- /dev/null
+++ b/tasks/bugs/validate-non-conflicting-ports-or-dirs.md
@@ -0,0 +1,10 @@
+---
+type: task
+---
+
+NetworkConfig should validate that no ports which have been configured to be
+used conflict with each other. In other words, no port used by a storage
+allocation should be used twice within that allocation, or used by another
+storage allocation.
+
+Same goes for directories being used.
diff --git a/tasks/misc/storage-allocation-status.md b/tasks/misc/storage-allocation-status.md
new file mode 100644
index 0000000..6620f7d
--- /dev/null
+++ b/tasks/misc/storage-allocation-status.md
@@ -0,0 +1,10 @@
+---
+type: task
+---
+
+There should be a `storage allocation status` command, separate from the
+`storage allocation list` command, which shows whether any allocations are
+currently draining (among any other information which might be useful).
+
+Once implemented the "Contributing Storage" operator documentation should be
+updated to recommend using this command during allocation removal.