[rqd] reserve all cores and report threads

VERSION.in

@@ -1 +1 @@
-1.8
+1.9

proto/report.proto

@@ -34,12 +34,13 @@ message BootReport {
 }
 
 message CoreDetail {
-    int32 total_cores = 1;
-    int32 idle_cores = 2;
-    int32 locked_cores = 3;
-    int32 booked_cores = 4;
+    int32 total_cores = 1; // physical cores
+    int32 total_threads = 2; // logical cores
+    int32 idle_cores = 3;
+    int32 locked_cores = 4;
+    int32 booked_cores = 5;
     //map <key: physid, value: coreid>
-    map<int64, CoreId> reserved_cores = 5;
+    map<int64, CoreId> reserved_cores = 6;
 }
 
 message CoreId {
@@ -67,21 +68,22 @@ message RenderHost {
     bool nimby_locked = 3; // if nimby has locked the host due to user activity
     string facility= 4; // The name of the facility that the host is in
     int32 num_procs = 5; // the number of physical procs on this machine
-    int32 cores_per_proc = 6; // the number of cores per proc
-    int64 total_swap = 7; // the total size of the swap in kB
-    int64 total_mem = 8; // the total size of the main memory pool in kB
-    int64 total_mcp = 9; // the total size of MCP in kB
-    int64 free_swap = 10; // the current amount of free swap in kB
-    int64 free_mem = 11; // the current amount of free memory in kB
-    int64 free_mcp = 12; // the current amount of free MCP in kB
-    int32 load = 13; // the current load on the proc
-    int32 boot_time = 14; // the time the proc was booted
-    repeated string tags = 15; // an array of default tags that are added to the host record
-    host.HardwareState state = 16; // hardware state for the host
-    map<string, string> attributes = 17; // additional data can be provided about the host
-    int32 num_gpus = 18; // the number of physical GPU's
-    int64 free_gpu_mem = 19; // the current amount of free gpu memory in kB
-    int64 total_gpu_mem = 20; // the total size of gpu memory in kB
+    int32 cores_per_proc = 6; // the number of physical cores per proc
+    int32 threads_per_proc = 7; // the number of logical cores per proc
+    int64 total_swap = 8; // the total size of the swap in kB
+    int64 total_mem = 9; // the total size of the main memory pool in kB
+    int64 total_mcp = 10; // the total size of MCP in kB
+    int64 free_swap = 11; // the current amount of free swap in kB
+    int64 free_mem = 12; // the current amount of free memory in kB
+    int64 free_mcp = 13; // the current amount of free MCP in kB
+    int32 load = 14; // the current load on the proc
+    int32 boot_time = 15; // the time the proc was booted
+    repeated string tags = 16; // an array of default tags that are added to the host record
+    host.HardwareState state = 17; // hardware state for the host
+    map<string, string> attributes = 18; // additional data can be provided about the host
+    int32 num_gpus = 19; // the number of physical GPU's
+    int64 free_gpu_mem = 20; // the current amount of free gpu memory in kB
+    int64 total_gpu_mem = 21; // the total size of gpu memory in kB
 };
 
 message RunningFrameInfo {
@@ -92,18 +94,19 @@ message RunningFrameInfo {
     string frame_name = 5;
     string layer_id = 6;
     int32 num_cores = 7;
-    int64 start_time = 8;
-    int64 max_rss = 9; // kB
-    int64 rss = 10; // kB
-    int64 max_vsize = 11; // kB
-    int64 vsize = 12; // kB
-    map<string, string> attributes = 13; //additional data can be provided about the running frame
-    int64 llu_time = 14;
-    int32 num_gpus = 15;
-    int64 max_used_gpu_memory = 16; // kB
-    int64 used_gpu_memory = 17; // kB
-    ChildrenProcStats children = 18; //additional data about the running frame's child processes
-    int64 used_swap_memory = 19; // kB
+    bool use_threads = 8; // whether the frame is requesting logical or physical cores
+    int64 start_time = 9;
+    int64 max_rss = 10; // kB
+    int64 rss = 11; // kB
+    int64 max_vsize = 12; // kB
+    int64 vsize = 13; // kB
+    map<string, string> attributes = 14; //additional data can be provided about the running frame
+    int64 llu_time = 15;
+    int32 num_gpus = 16;
+    int64 max_used_gpu_memory = 17; // kB
+    int64 used_gpu_memory = 18; // kB
+    ChildrenProcStats children = 19; //additional data about the running frame's child processes
+    int64 used_swap_memory = 20; // kB
 };
 
 message ChildrenProcStats {

rqd/rqd/__main__.py

@@ -31,6 +31,7 @@
 In /etc/opencue/rqd.conf (on Linux) or %LOCALAPPDATA%/OpenCue/rqd.conf (on Windows):
 [Override]
 OVERRIDE_CORES = 2
+OVERRIDE_THREADS = 4
 OVERRIDE_PROCS = 3
 OVERRIDE_MEMORY = 1000000
 OVERRIDE_CUEBOT = cuebot1 cuebot2 cuebot3

rqd/rqd/rqconstants.py

@@ -150,15 +150,16 @@
 if '-c' in sys.argv:
     CONFIG_FILE = sys.argv[sys.argv.index('-c') + 1]
 
-OVERRIDE_CORES = None # number of cores. ex: None or 8
-OVERRIDE_IS_DESKTOP = None # Force rqd to run in 'desktop' mode
 OVERRIDE_PROCS = None # number of physical cpus. ex: None or 2
+OVERRIDE_CORES = None # number of cores per cpu. ex: None or 8
+OVERRIDE_THREADS = None # number of threads per cpu. ex: None or 16
+OVERRIDE_IS_DESKTOP = None # Force rqd to run in 'desktop' mode
 OVERRIDE_MEMORY = None # in Kb
 OVERRIDE_NIMBY = None # True to turn on, False to turn off
 USE_NIMBY_PYNPUT = True # True pynput, False select
 OVERRIDE_HOSTNAME = None # Force to use this hostname
 ALLOW_GPU = False
-LOAD_MODIFIER = 0 # amount to add/subtract from load
+LOAD_MODIFIER = 0 # amount to add/subtract from load,
 
 LOG_FORMAT = '%(levelname)-9s openrqd-%(module)-10s: %(message)s'
 CONSOLE_LOG_LEVEL = logging.WARNING
@@ -203,6 +204,8 @@
             CUEBOT_GRPC_PORT = config.getint(__override_section, "CUEBOT_GRPC_PORT")
         if config.has_option(__override_section, "OVERRIDE_CORES"):
             OVERRIDE_CORES = config.getint(__override_section, "OVERRIDE_CORES")
+        if config.has_option(__override_section, "OVERRIDE_THREADS"):
+            OVERRIDE_THREADS = config.getint(__override_section, "OVERRIDE_THREADS")
         if config.has_option(__override_section, "OVERRIDE_PROCS"):
             OVERRIDE_PROCS = config.getint(__override_section, "OVERRIDE_PROCS")
         if config.has_option(__override_section, "OVERRIDE_MEMORY"):
@@ -230,7 +233,7 @@
             RQD_USE_PATH_ENV_VAR = config.getboolean(__override_section, "RQD_USE_PATH_ENV_VAR")
         if config.has_option(__override_section, "RQD_USE_ALL_HOST_ENV_VARS"):
             RQD_USE_HOST_ENV_VARS = config.getboolean(__override_section,
-                "RQD_USE_ALL_HOST_ENV_VARS")
+                                                      "RQD_USE_ALL_HOST_ENV_VARS")
         if config.has_option(__override_section, "RQD_BECOME_JOB_USER"):
             RQD_BECOME_JOB_USER = config.getboolean(__override_section, "RQD_BECOME_JOB_USER")
         if config.has_option(__override_section, "RQD_TAGS"):

rqd/rqd/rqcore.py

@@ -68,6 +68,7 @@ def __init__(self, optNimbyoff=False):
 
         self.cores = rqd.compiled_proto.report_pb2.CoreDetail(
             total_cores=0,
+            total_threads=0,
             idle_cores=0,
             locked_cores=0,
             booked_cores=0,
@@ -317,7 +318,10 @@ def killAllFrame(self, reason):
     def releaseCores(self, reqRelease, releaseHT=None, releaseGpus=None):
         """The requested number of cores are released
         @type  reqRelease: int
-        @param reqRelease: Number of cores to release, 100 = 1 physical core"""
+        @param reqRelease: Number of cores to release, 100 = 1 physical core
+        @type releaseHT: str
+        @param releaseHT: The hyper-threading cores to release
+        """
         with self.__threadLock:
             # pylint: disable=no-member
             self.cores.booked_cores -= reqRelease
@@ -420,9 +424,11 @@ def launchFrame(self, runFrame):
             # pylint: enable=no-member
 
             if runFrame.environment.get('CUE_THREADABLE') == '1':
-                reserveHT = self.machine.reserveHT(runFrame.num_cores)
-                if reserveHT:
-                    runFrame.attributes['CPU_LIST'] = reserveHT
+                cpu_list = self.machine.reserveCores(
+                    coresCount=runFrame.num_cores,
+                    logical=runFrame.use_threads)
+                if cpu_list:
+                    runFrame.attributes['CPU_LIST'] = cpu_list
 
             if runFrame.num_gpus:
                 reserveGpus = self.machine.reserveGpus(runFrame.num_gpus)