feat: implement resource validation checks for production deployment

- Added a new ResourceChecker type to validate system resources including disk space, RAM, and CPU cores. - Implemented CheckDiskSpace, CheckRAM, and CheckCPU methods to ensure minimum requirements are met for production deployment. - Integrated resource checks into the ProductionSetup's Phase1CheckPrerequisites method to enhance deployment reliability. - Updated systemd service generation to log output to specific log files instead of the journal for better log management.
2025-12-11 09:18:50 +00:00 · 2025-11-11 05:44:40 +02:00 · 2025-11-11 05:44:40 +02:00 · e9bf94ba96
commit e9bf94ba96
parent 52a726ffd4
5 changed files with 147 additions and 21 deletions
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@ -13,6 +13,21 @@ The format is based on [Keep a Changelog][keepachangelog] and adheres to [Semant
 ### Deprecated

 ### Fixed
+## [0.66.0] - 2025-11-11
+
+### Added
+- Pre-installation checks for minimum system resources (10GB disk space, 2GB RAM, 2 CPU cores) are now performed during setup.
+- All systemd services (IPFS, RQLite, Olric, Node, Gateway) now log directly to dedicated files in the logs directory instead of using the system journal.
+
+### Changed
+- Improved logging instructions in the setup completion message to reference the new dedicated log files.
+
+### Deprecated
+
+### Removed
+
+### Fixed
+\n
 ## [0.65.0] - 2025-11-11

 ### Added
--- a/2
+++ b/2
@ -19,7 +19,7 @@ test-e2e:

 .PHONY: build clean test run-node run-node2 run-node3 run-example deps tidy fmt vet lint clear-ports install-hooks kill

-VERSION := 0.65.0
+VERSION := 0.66.0
 COMMIT  ?= $(shell git rev-parse --short HEAD 2>/dev/null || echo unknown)
 DATE    ?= $(shell date -u +%Y-%m-%dT%H:%M:%SZ)
 LDFLAGS := -X 'main.version=$(VERSION)' -X 'main.commit=$(COMMIT)' -X 'main.date=$(DATE)'
--- a/pkg/environments/production/checks.go
+++ b/pkg/environments/production/checks.go
@ -5,7 +5,9 @@ import (
 	"os"
 	"os/exec"
 	"runtime"
+	"strconv"
 	"strings"
+	"syscall"
 )

 // OSInfo contains detected operating system information
@ -213,3 +215,74 @@ func (etc *ExternalToolChecker) CheckGoAvailable() bool {
 	_, err := exec.LookPath("go")
 	return err == nil
 }
+
+// ResourceChecker validates system resources for production deployment
+type ResourceChecker struct{}
+
+// NewResourceChecker creates a new resource checker
+func NewResourceChecker() *ResourceChecker {
+	return &ResourceChecker{}
+}
+
+// CheckDiskSpace validates sufficient disk space (minimum 10GB free)
+func (rc *ResourceChecker) CheckDiskSpace(path string) error {
+	var stat syscall.Statfs_t
+	if err := syscall.Statfs(path, &stat); err != nil {
+		return fmt.Errorf("failed to check disk space: %w", err)
+	}
+
+	// Available space in bytes
+	availableBytes := stat.Bavail * uint64(stat.Bsize)
+	minRequiredBytes := uint64(10 * 1024 * 1024 * 1024) // 10GB
+
+	if availableBytes < minRequiredBytes {
+		availableGB := float64(availableBytes) / (1024 * 1024 * 1024)
+		return fmt.Errorf("insufficient disk space: %.1fGB available, minimum 10GB required", availableGB)
+	}
+
+	return nil
+}
+
+// CheckRAM validates sufficient RAM (minimum 2GB total)
+func (rc *ResourceChecker) CheckRAM() error {
+	data, err := os.ReadFile("/proc/meminfo")
+	if err != nil {
+		return fmt.Errorf("failed to read memory info: %w", err)
+	}
+
+	lines := strings.Split(string(data), "\n")
+	totalKB := uint64(0)
+
+	for _, line := range lines {
+		if strings.HasPrefix(line, "MemTotal:") {
+			parts := strings.Fields(line)
+			if len(parts) >= 2 {
+				if kb, err := strconv.ParseUint(parts[1], 10, 64); err == nil {
+					totalKB = kb
+					break
+				}
+			}
+		}
+	}
+
+	if totalKB == 0 {
+		return fmt.Errorf("could not determine total RAM")
+	}
+
+	minRequiredKB := uint64(2 * 1024 * 1024) // 2GB in KB
+	if totalKB < minRequiredKB {
+		totalGB := float64(totalKB) / (1024 * 1024)
+		return fmt.Errorf("insufficient RAM: %.1fGB total, minimum 2GB required", totalGB)
+	}
+
+	return nil
+}
+
+// CheckCPU validates sufficient CPU cores (minimum 2 cores)
+func (rc *ResourceChecker) CheckCPU() error {
+	cores := runtime.NumCPU()
+	if cores < 2 {
+		return fmt.Errorf("insufficient CPU cores: %d available, minimum 2 required", cores)
+	}
+	return nil
+}
--- a/pkg/environments/production/orchestrator.go
+++ b/pkg/environments/production/orchestrator.go
@ -21,6 +21,7 @@ type ProductionSetup struct {
 	privChecker       *PrivilegeChecker
 	osDetector        *OSDetector
 	archDetector      *ArchitectureDetector
+	resourceChecker   *ResourceChecker
 	fsProvisioner     *FilesystemProvisioner
 	userProvisioner   *UserProvisioner
 	stateDetector     *StateDetector
@ -48,6 +49,7 @@ func NewProductionSetup(debrosHome string, logWriter io.Writer, forceReconfigure
 		privChecker:       &PrivilegeChecker{},
 		osDetector:        &OSDetector{},
 		archDetector:      &ArchitectureDetector{},
+		resourceChecker:   NewResourceChecker(),
 		fsProvisioner:     NewFilesystemProvisioner(debrosHome),
 		userProvisioner:   NewUserProvisioner("debros", debrosHome, "/bin/bash"),
 		stateDetector:     NewStateDetector(debrosDir),
@ -115,6 +117,25 @@ func (ps *ProductionSetup) Phase1CheckPrerequisites() error {
 	}
 	ps.logf("  ✓ Basic dependencies available")

+	// Check system resources
+	if err := ps.resourceChecker.CheckDiskSpace(ps.debrosHome); err != nil {
+		ps.logf("  ❌ %v", err)
+		return err
+	}
+	ps.logf("  ✓ Sufficient disk space available")
+
+	if err := ps.resourceChecker.CheckRAM(); err != nil {
+		ps.logf("  ❌ %v", err)
+		return err
+	}
+	ps.logf("  ✓ Sufficient RAM available")
+
+	if err := ps.resourceChecker.CheckCPU(); err != nil {
+		ps.logf("  ❌ %v", err)
+		return err
+	}
+	ps.logf("  ✓ Sufficient CPU cores available")
+
 	return nil
 }

@ -426,8 +447,15 @@ func (ps *ProductionSetup) LogSetupComplete(peerID string) {
 	ps.logf("\nNode Peer ID: %s", peerID)
 	ps.logf("\nService Management:")
 	ps.logf("  systemctl status debros-ipfs-bootstrap")
-	ps.logf("  systemctl logs debros-node-bootstrap")
-	ps.logf("  sudo tail -f %s/logs/node.log", ps.debrosDir)
+	ps.logf("  journalctl -u debros-node-bootstrap -f")
+	ps.logf("  tail -f %s/logs/node-bootstrap.log", ps.debrosDir)
+	ps.logf("\nLog Files:")
+	ps.logf("  %s/logs/ipfs-bootstrap.log", ps.debrosDir)
+	ps.logf("  %s/logs/ipfs-cluster-bootstrap.log", ps.debrosDir)
+	ps.logf("  %s/logs/rqlite-bootstrap.log", ps.debrosDir)
+	ps.logf("  %s/logs/olric.log", ps.debrosDir)
+	ps.logf("  %s/logs/node-bootstrap.log", ps.debrosDir)
+	ps.logf("  %s/logs/gateway.log", ps.debrosDir)
 	ps.logf("\nStart All Services:")
 	ps.logf("  systemctl start debros-ipfs-bootstrap debros-ipfs-cluster-bootstrap debros-rqlite-bootstrap debros-olric debros-node-bootstrap debros-gateway")
 	ps.logf("\nVerify Installation:")
--- a/pkg/environments/production/services.go
+++ b/pkg/environments/production/services.go
@ -31,6 +31,8 @@ func (ssg *SystemdServiceGenerator) GenerateIPFSService(nodeType string) string
 		ipfsRepoPath = filepath.Join(ssg.debrosDir, "data", "node", "ipfs", "repo")
 	}

+	logFile := filepath.Join(ssg.debrosDir, "logs", fmt.Sprintf("ipfs-%s.log", nodeType))
+
 	return fmt.Sprintf(`[Unit]
 Description=IPFS Daemon (%s)
 After=network-online.target
@ -46,8 +48,8 @@ ExecStartPre=/bin/bash -c 'if [ -f %s/secrets/swarm.key ] && [ ! -f %s/swarm.key
 ExecStart=/usr/bin/ipfs daemon --enable-pubsub-experiment --repo-dir=%s
 Restart=always
 RestartSec=5
-StandardOutput=journal
-StandardError=journal
+StandardOutput=file:%s
+StandardError=file:%s
 SyslogIdentifier=ipfs-%s

 NoNewPrivileges=yes
@ -57,7 +59,7 @@ ReadWritePaths=%s

 [Install]
 WantedBy=multi-user.target
-`, nodeType, ssg.debrosHome, ipfsRepoPath, ssg.debrosDir, ipfsRepoPath, ssg.debrosDir, ipfsRepoPath, ipfsRepoPath, ipfsRepoPath, nodeType, ssg.debrosDir)
+`, nodeType, ssg.debrosHome, ipfsRepoPath, ssg.debrosDir, ipfsRepoPath, ssg.debrosDir, ipfsRepoPath, ipfsRepoPath, ipfsRepoPath, logFile, logFile, nodeType, ssg.debrosDir)
 }

 // GenerateIPFSClusterService generates the IPFS Cluster systemd unit
@ -69,6 +71,8 @@ func (ssg *SystemdServiceGenerator) GenerateIPFSClusterService(nodeType string)
 		clusterPath = filepath.Join(ssg.debrosDir, "data", "node", "ipfs-cluster")
 	}

+	logFile := filepath.Join(ssg.debrosDir, "logs", fmt.Sprintf("ipfs-cluster-%s.log", nodeType))
+
 	return fmt.Sprintf(`[Unit]
 Description=IPFS Cluster Service (%s)
 After=debros-ipfs-%s.service
@ -85,8 +89,8 @@ Environment=IPFS_CLUSTER_PATH=%s
 ExecStart=/usr/local/bin/ipfs-cluster-service daemon
 Restart=always
 RestartSec=5
-StandardOutput=journal
-StandardError=journal
+StandardOutput=file:%s
+StandardError=file:%s
 SyslogIdentifier=ipfs-cluster-%s

 NoNewPrivileges=yes
@ -96,7 +100,7 @@ ReadWritePaths=%s

 [Install]
 WantedBy=multi-user.target
-`, nodeType, nodeType, nodeType, nodeType, ssg.debrosHome, ssg.debrosHome, clusterPath, nodeType, ssg.debrosDir)
+`, nodeType, nodeType, nodeType, nodeType, ssg.debrosHome, ssg.debrosHome, clusterPath, logFile, logFile, nodeType, ssg.debrosDir)
 }

 // GenerateRQLiteService generates the RQLite systemd unit
@ -119,6 +123,8 @@ func (ssg *SystemdServiceGenerator) GenerateRQLiteService(nodeType string, httpP

 	args += fmt.Sprintf(` %s`, dataDir)

+	logFile := filepath.Join(ssg.debrosDir, "logs", fmt.Sprintf("rqlite-%s.log", nodeType))
+
 	return fmt.Sprintf(`[Unit]
 Description=RQLite Database (%s)
 After=network-online.target
@ -132,8 +138,8 @@ Environment=HOME=%s
 ExecStart=/usr/local/bin/rqlited %s
 Restart=always
 RestartSec=5
-StandardOutput=journal
-StandardError=journal
+StandardOutput=file:%s
+StandardError=file:%s
 SyslogIdentifier=rqlite-%s

 NoNewPrivileges=yes
@ -143,12 +149,13 @@ ReadWritePaths=%s

 [Install]
 WantedBy=multi-user.target
-`, nodeType, ssg.debrosHome, args, nodeType, ssg.debrosDir)
+`, nodeType, ssg.debrosHome, args, logFile, logFile, nodeType, ssg.debrosDir)
 }

 // GenerateOlricService generates the Olric systemd unit
 func (ssg *SystemdServiceGenerator) GenerateOlricService() string {
 	olricConfigPath := filepath.Join(ssg.debrosDir, "configs", "olric", "config.yaml")
+	logFile := filepath.Join(ssg.debrosDir, "logs", "olric.log")

 	return fmt.Sprintf(`[Unit]
 Description=Olric Cache Server
@ -164,8 +171,8 @@ Environment=OLRIC_SERVER_CONFIG=%s
 ExecStart=/usr/local/bin/olric-server
 Restart=always
 RestartSec=5
-StandardOutput=journal
-StandardError=journal
+StandardOutput=file:%s
+StandardError=file:%s
 SyslogIdentifier=olric

 NoNewPrivileges=yes
@ -175,7 +182,7 @@ ReadWritePaths=%s

 [Install]
 WantedBy=multi-user.target
-`, ssg.debrosHome, olricConfigPath, ssg.debrosDir)
+`, ssg.debrosHome, olricConfigPath, logFile, logFile, ssg.debrosDir)
 }

 // GenerateNodeService generates the DeBros Node systemd unit
@ -187,6 +194,8 @@ func (ssg *SystemdServiceGenerator) GenerateNodeService(nodeType string) string
 		configFile = "node.yaml"
 	}

+	logFile := filepath.Join(ssg.debrosDir, "logs", fmt.Sprintf("node-%s.log", nodeType))
+
 	return fmt.Sprintf(`[Unit]
 Description=DeBros Network Node (%s)
 After=debros-ipfs-cluster-%s.service
@ -202,8 +211,8 @@ Environment=HOME=%s
 ExecStart=%s/bin/node --config %s/configs/%s
 Restart=always
 RestartSec=5
-StandardOutput=journal
-StandardError=journal
+StandardOutput=file:%s
+StandardError=file:%s
 SyslogIdentifier=debros-node-%s

 NoNewPrivileges=yes
@ -213,12 +222,13 @@ ReadWritePaths=%s

 [Install]
 WantedBy=multi-user.target
-`, nodeType, nodeType, nodeType, nodeType, ssg.debrosHome, ssg.debrosHome, ssg.debrosHome, ssg.debrosDir, configFile, nodeType, ssg.debrosDir)
+`, nodeType, nodeType, nodeType, nodeType, ssg.debrosHome, ssg.debrosHome, ssg.debrosHome, ssg.debrosDir, configFile, logFile, logFile, nodeType, ssg.debrosDir)
 }

 // GenerateGatewayService generates the DeBros Gateway systemd unit
 func (ssg *SystemdServiceGenerator) GenerateGatewayService(nodeType string) string {
 	nodeService := fmt.Sprintf("debros-node-%s.service", nodeType)
+	logFile := filepath.Join(ssg.debrosDir, "logs", "gateway.log")
 	return fmt.Sprintf(`[Unit]
 Description=DeBros Gateway
 After=%s
@ -233,8 +243,8 @@ Environment=HOME=%s
 ExecStart=%s/bin/gateway --config %s/configs/gateway.yaml
 Restart=always
 RestartSec=5
-StandardOutput=journal
-StandardError=journal
+StandardOutput=file:%s
+StandardError=file:%s
 SyslogIdentifier=debros-gateway

 AmbientCapabilities=CAP_NET_BIND_SERVICE
@ -247,7 +257,7 @@ ReadWritePaths=%s

 [Install]
 WantedBy=multi-user.target
-`, nodeService, nodeService, ssg.debrosHome, ssg.debrosHome, ssg.debrosHome, ssg.debrosDir, ssg.debrosDir)
+`, nodeService, nodeService, ssg.debrosHome, ssg.debrosHome, ssg.debrosHome, ssg.debrosDir, logFile, logFile, ssg.debrosDir)
 }

 // SystemdController manages systemd service operations