idrac 0.10.3 → 0.10.5

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 67737e81c936f047dc3264470e8cf3d913e10d0b300af332f4bd16aca03f4078
-  data.tar.gz: 62467838651fb0d17ab870948d1b3a82e02f5de31297a2e6d7d7ed260dea044a
+  metadata.gz: 9c4504dd010f1458b9705e73489d4d68ec18214ef13a5a5187dde3ed52812d1a
+  data.tar.gz: eb9ab621e77b1ad17ade9579b4286a7696f7f0bb9dc0761e8862446ee8f39963
 SHA512:
-  metadata.gz: 4164655c54761c104a7b43f5f489972d74238c7f086b3bd865988e875eb756edaa751b6bd3241b0655c154fae57f575849d09796bbc9c0feca015192def2cbe3
-  data.tar.gz: 5aff6a5a4cbd3362e6dab402f493759ae7b0db72f75adc40c20a9cc84da2e1edea8990d27cd3b0dd4f10ee21ed0775404d101a2dd09752a8a55883dcc2312303
+  metadata.gz: af0824a2ceb96d5c89a2552277d24bdf754eb748a7644d756deb3864620130b005d3f5cefa1f0921d6617fd03aab79c31fd8d9de8b0716a2e3124a8916f1bde4
+  data.tar.gz: 3c85876b2e93b31bed8f8dd4a8ea2ab4ae851048a2767c18e319151b74748c797502971c0932b9e534150c95ce66c3bb9ff93e5194a0d6e37108714724363396

data/lib/idrac/boot.rb

@@ -323,22 +323,7 @@ module IDRAC
         raise Error, "Failed to get BIOS information. Status code: #{response.status}"
       end
     end
-=begin
-    # Servers can boot in BIOS mode or in UEFI (modern, extensible BIOS replacement) mode.
-    # We use UEFI mode.
-    # self.get(path: "Systems/System.Embedded.1/Bios/Settings?$select=BootMode")
-    res = self.get(path: "Systems/System.Embedded.1/Bios")
-    if res["body"]["Attributes"]["BootMode"] == "Uefi"
-      return { status: :success }
-    else
-      res = self.set_system_configuration_profile(scp_boot_mode_uefi, reboot: true)
-      # Then must power cycle the server
-      self.power_on!(wait: true)
-      self.power_off!(wait: true)
-      return res
-    end
 
-=end
     def scp_boot_mode_uefi(idrac_license_version: 9)
       opts = { "BootMode" => 'Uefi' }
       # If we're iDRAC 9, we need enable a placeholder, otherwise we can't order the

data/lib/idrac/firmware.rb

@@ -90,12 +90,29 @@ module IDRAC
               # Id ("Installed-<componentID>-<version>"). Used to match DUPs by
               # componentID rather than by ambiguous display name.
               fw_id = component_data['Id'].to_s
-              component_id = component_data.dig('Oem', 'Dell', 'DellSoftwareInventory', 'ComponentID')
+              oem = component_data.dig('Oem', 'Dell', 'DellSoftwareInventory') || {}
+              component_id = oem['ComponentID']
               component_id ||= fw_id[/\A(?:Installed|Current|Previous|Available)-(\d+)-/, 1]
+              # The same Oem block carries the device's PCI IDs, and it carries
+              # them exactly where the componentID is missing. The iDRAC reports
+              # componentID "0" for firmware it did not install itself, so
+              # factory-flashed hardware is unmatchable by componentID: on an
+              # R6525, 13 of 23 entries report "0" (both LOMs, both ConnectX-6,
+              # the PERC, all 8 NVMe drives) and each publishes all four PCI
+              # IDs, while the 10 with a real componentID publish none. An
+              # entry gains its componentID once a DUP is applied — a LOM went
+              # from "Installed-0-22.91.5" to "Installed-108255-23.61.3" — so
+              # the unmatchable set is first-touch hardware, not an edge case.
+              # No PCIeDevices/PCIeFunctions lookup is needed to identify what
+              # the componentID cannot; see #updates_for_component.
               firmware_inventory << {
                 name: component_data['Name'],
                 id: component_data['Id'],
                 component_id: component_id,
+                vendor_id: oem['VendorID'],
+                device_id: oem['DeviceID'],
+                sub_vendor_id: oem['SubVendorID'],
+                sub_device_id: oem['SubDeviceID'],
                 version: component_data['Version'],
                 updateable: component_data['Updateable'] || false,
                 status: component_data['Status'] ? component_data['Status']['State'] : 'Unknown'

data/lib/idrac/firmware_catalog.rb

@@ -172,6 +172,16 @@ module IDRAC
         # picks wrong DUPs and iDRAC rejects them (RED097 / "not compatible").
         component_ids = component.xpath(".//SupportedDevices/Device/@componentID").map(&:value).uniq
 
+        # The PCI IDs this DUP declares it can flash. The PCIe/storage half of
+        # the iDRAC inventory reports componentID "0" (measured on an R6525:
+        # every NIC, the PERC and all 8 NVMe drives), so componentID alone
+        # cannot match them and the old name heuristic guessed — it offered a
+        # Broadcom LOM an Intel package. PCI IDs are the identity both sides
+        # publish, so a cross-vendor match becomes impossible.
+        pci_ids = component.xpath(".//SupportedDevices/Device/PCIInfo").map { |p|
+          pci_key(p["vendorID"], p["deviceID"], p["subVendorID"], p["subDeviceID"])
+        }.compact.uniq
+
         # Skip if missing essential information
         next if name.empty? || path.empty? || version.empty?
 
@@ -191,6 +201,7 @@ module IDRAC
             component_type: component_type,
             category: category,
             component_ids: component_ids,
+            pci_ids: pci_ids,
             download_url: "https://downloads.dell.com/#{path}"
           }
         end
@@ -200,24 +211,66 @@ module IDRAC
       updates
     end
 
+    # Normalize one PCI ID quad to a comparable key. Dell publishes the same
+    # four IDs on both sides but spells them differently: the catalog writes
+    # bare uppercase hex (<PCIInfo vendorID="14E4" deviceID="165F"/>) while
+    # Redfish writes "0x790e". Returns nil unless all four are present, so a
+    # caller can tell "no PCI identity" from "PCI identity that matched
+    # nothing".
+    def pci_key(vendor_id, device_id, sub_vendor_id, sub_device_id)
+      parts = [vendor_id, device_id, sub_vendor_id, sub_device_id].map do |id|
+        id.to_s.strip.sub(/\A0x/i, "").upcase
+      end
+      return nil if parts.any?(&:empty?)
+
+      parts.map { |p| p.rjust(4, "0") }.join("/")
+    end
+
     # Select the catalog DUPs that apply to one installed component (`fw` from
-    # the iDRAC inventory). Matches on Dell componentID — the reliable key —
-    # and only falls back to the legacy name heuristic when the inventory
-    # didn't expose a componentID (e.g. it reports "0").
+    # the iDRAC inventory), most specific key first.
+    #
+    # 1. Dell componentID, when the inventory gives a real one. Exact and
+    #    unambiguous; unchanged.
+    # 2. PCI IDs, for the entries that report componentID "0" — on an R6525
+    #    that is every NIC, the PERC and every drive. The full quad first, then
+    #    vendor+device if the catalog lists no package for that exact subsystem
+    #    ID. Both tiers pin the vendor, which is the whole point: name matching
+    #    reduced any NIC to the token "NIC" and any RAID device to "PERC", then
+    #    substring-matched the DUP's package name, so a "Broadcom NetXtreme
+    #    Gigabit Ethernet" LOM was offered "Intel NIC Family Version 25.0.0
+    #    Firmware for Intel I350 and X550 Adapters" — wrong vendor — and a
+    #    "PERC H755N Front" was offered the H355 package. It was also unstable:
+    #    the same tool offered two identical PERCs different packages, because
+    #    the winner depended on catalog order.
+    # 3. The name heuristic, only when the inventory gave neither key. When PCI
+    #    IDs are present they are the answer: if nothing matches them, the
+    #    catalog has no package for this device, and guessing by name is how
+    #    the wrong-vendor DUPs got offered in the first place.
     def updates_for_component(catalog_updates, fw)
       component_id = fw[:component_id]
       if component_id && component_id != "0"
-        catalog_updates.select { |u| Array(u[:component_ids]).include?(component_id) }
-      else
-        name = fw[:name] || ""
-        ids = extract_identifiers(name)
-        catalog_updates.select do |u|
-          un = u[:name] || ""
-          ids.any? { |id| un.downcase.include?(id.downcase) } ||
-            un.downcase.include?(name.downcase) ||
-            name.downcase.include?(un.downcase)
+        return catalog_updates.select { |u| Array(u[:component_ids]).include?(component_id) }
+      end
+
+      key = pci_key(fw[:vendor_id], fw[:device_id], fw[:sub_vendor_id], fw[:sub_device_id])
+      if key
+        exact = catalog_updates.select { |u| Array(u[:pci_ids]).include?(key) }
+        return exact if exact.any?
+
+        chip = key.split("/").first(2).join("/")
+        return catalog_updates.select do |u|
+          Array(u[:pci_ids]).any? { |k| k.start_with?("#{chip}/") }
         end
       end
+
+      name = fw[:name] || ""
+      ids = extract_identifiers(name)
+      catalog_updates.select do |u|
+        un = u[:name] || ""
+        ids.any? { |id| un.downcase.include?(id.downcase) } ||
+          un.downcase.include?(name.downcase) ||
+          name.downcase.include?(un.downcase)
+      end
     end
 
     # Sortable key for a Dell version string (e.g. "25.5.9.0001", "22.00.6").

data/lib/idrac/power.rb

@@ -58,7 +58,15 @@ module IDRAC
       return true
     end
     
+    # Reset types that restart the host rather than leave it down. A restart never settles on
+    # "Off", so it must never be waited on — or escalated to — as if it were a shutdown.
+    RESTART_TYPES = %w[ForceRestart GracefulRestart PowerCycle].freeze
+
     def power_off(wait: true, kind: "ForceOff")
+      # A restart is not a shutdown. Sent down the power-off path it fails the wait for "Off" and
+      # escalates to ForceOff, which leaves the host powered down. Route it to the restart instead.
+      return reboot(kind: kind, wait: wait) if RESTART_TYPES.include?(kind)
+
       ensure_authenticated!
       
       puts "Powering off server...".light_cyan
@@ -93,37 +101,39 @@ module IDRAC
       return true
     end
     
-    def reboot
+    def reboot(kind: "ForceRestart", wait: false)
       ensure_authenticated!
       
-      puts "Rebooting server...".light_cyan
+      puts "Rebooting server (#{kind})...".light_cyan
 
       # Check current power state first
       current_state = get_power_state rescue "Unknown"
       if current_state == "Off"
         puts "Server is currently off, powering on instead of rebooting".yellow
-        return power_on
+        return power_on(wait: wait)
       end
       
-      # Send reboot command (Reset with ResetType=ForceRestart)
       path = "/redfish/v1/Systems/System.Embedded.1/Actions/ComputerSystem.Reset"
-      payload = { "ResetType" => "ForceRestart" }
-      
-      response = authenticated_request(:post, path, body: payload.to_json)
+      response = authenticated_request(:post, path, body: { "ResetType" => kind }.to_json)
       
       if response.status >= 200 && response.status < 300
         puts "Server reboot command sent successfully".green
-        return true
-      elsif response.status == 409
+      elsif response.status == 409 && kind != "GracefulRestart"
         error_data = JSON.parse(response.body) rescue nil
         puts "Received conflict (409) error from iDRAC: #{error_data.inspect}"
         # Try gracefulRestart as an alternative
         puts "Trying GracefulRestart instead...".yellow
-        payload = { "ResetType" => "GracefulRestart" }
-        authenticated_request(:post, path, body: payload.to_json)
+        response = authenticated_request(:post, path, body: { "ResetType" => "GracefulRestart" }.to_json)
+        raise Error, "Failed to reboot server. Status code: #{response.status}" unless response.status.between?(200, 299)
       else
         raise Error, "Failed to reboot server. Status code: #{response.status}"
       end
+      
+      # "On" is the only terminal state a restart reaches. Never wait for "Off" here: the host may
+      # never report it, and treating that as failure is what turned a reboot into a shutdown.
+      wait_for_power_state(target_state: "On", tries: 12) if wait
+      
+      return true
     end
     
     def get_power_state

data/lib/idrac/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module IDRAC
-  VERSION = "0.10.3"
+  VERSION = "0.10.5"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: idrac
 version: !ruby/object:Gem::Version
-  version: 0.10.3
+  version: 0.10.5
 platform: ruby
 authors:
 - Jonathan Siegel