Merge pull request #45182 from fastmachinelearning/cmsTritonUpdates

SONIC updates for site support

Author: cmsbuild

HeterogeneousCore/SonicTriton/README.md

@@ -132,19 +132,19 @@ The script has three operations (`start`, `stop`, `check`) and the following opt
 * `-c`: don't cleanup temporary dir (for debugging)
 * `-C [dir]`: directory containing Nvidia compatibility drivers (checks CMSSW_BASE by default if available)
 * `-D`: dry run: print container commands rather than executing them
-* `-d`: use Docker instead of Apptainer
+* `-d [exe]`: container choice: apptainer, docker, podman, podman-hpc (default: apptainer)
 * `-E [path]`: include extra path(s) for executables (default: /cvmfs/oasis.opensciencegrid.org/mis/apptainer/current/bin)
 * `-f`: force reuse of (possibly) existing container instance
-* `-g`: use GPU instead of CPU
-* `-i` [name]`: server image name (default: fastml/triton-torchgeo:22.07-py3-geometric)
+* `-g [device]`: device choice: auto (try to detect GPU), CPU, GPU (default: auto)
+* `-i [name]`: server image name (default: fastml/triton-torchgeo:22.07-py3-geometric)
 * `-I [num]`: number of model instances (default: 0 -> means no local editing of config files)
 * `-M [dir]`: model repository (can be given more than once)
 * `-m [dir]`: specific model directory (can be given more than one)
 * `-n [name]`: name of container instance, also used for hidden temporary dir (default: triton_server_instance)
 * `-P [port]`: base port number for services (-1: automatically find an unused port range) (default: 8000)
 * `-p [pid]`: automatically shut down server when process w/ specified PID ends (-1: use parent process PID)
 * `-r [num]`: number of retries when starting container (default: 3)
-* `-s [dir]`: Apptainer sandbox directory (default: /cvmfs/unpacked.cern.ch/registry.hub.docker.com/fastml/triton-torchgeo:22.07-py3-geometric)
+* `-s [dir]`: apptainer sandbox directory (default: /cvmfs/unpacked.cern.ch/registry.hub.docker.com/fastml/triton-torchgeo:22.07-py3-geometric)
 * `-t [dir]`: non-default hidden temporary dir
 * `-v`: (verbose) start: activate server debugging info; stop: keep server logs
 * `-w [time]`: maximum time to wait for server to start (default: 300 seconds)
@@ -200,8 +200,8 @@ The fallback server has a separate set of options, mostly related to the invocat
 * `enable`: enable the fallback server
 * `debug`: enable debugging (equivalent to `-c` in `cmsTriton`)
 * `verbose`: enable verbose output in logs (equivalent to `-v` in `cmsTriton`)
-* `useDocker`: use Docker instead of Apptainer (equivalent to `-d` in `cmsTriton`)
-* `useGPU`: run on local GPU (equivalent to `-g` in `cmsTriton`)
+* `container`: container choice (equivalent to `-d` in `cmsTriton`)
+* `device`: device choice (equivalent to `-g` in `cmsTriton`)
 * `retries`: number of retries when starting container (passed to `-r [num]` in `cmsTriton` if >= 0; default: -1)
 * `wait`: maximum time to wait for server to start (passed to `-w time` in `cmsTriton` if >= 0; default: -1)
 * `instanceBaseName`: base name for server instance if random names are enabled (default: triton_server_instance)

HeterogeneousCore/SonicTriton/interface/TritonService.h

@@ -36,8 +36,8 @@ class TritonService {
         : enable(pset.getUntrackedParameter<bool>("enable")),
           debug(pset.getUntrackedParameter<bool>("debug")),
           verbose(pset.getUntrackedParameter<bool>("verbose")),
-          useDocker(pset.getUntrackedParameter<bool>("useDocker")),
-          useGPU(pset.getUntrackedParameter<bool>("useGPU")),
+          container(pset.getUntrackedParameter<std::string>("container")),
+          device(pset.getUntrackedParameter<std::string>("device")),
           retries(pset.getUntrackedParameter<int>("retries")),
           wait(pset.getUntrackedParameter<int>("wait")),
           instanceName(pset.getUntrackedParameter<std::string>("instanceName")),
@@ -54,8 +54,8 @@ class TritonService {
     bool enable;
     bool debug;
     bool verbose;
-    bool useDocker;
-    bool useGPU;
+    std::string container;
+    std::string device;
     int retries;
     int wait;
     std::string instanceName;
@@ -89,6 +89,7 @@ class TritonService {
     std::unordered_set<std::string> models;
     static const std::string fallbackName;
     static const std::string fallbackAddress;
+    static const std::string siteconfName;
   };
   struct Model {
     Model(const std::string& path_ = "") : path(path_) {}

HeterogeneousCore/SonicTriton/python/TritonService_cff.py

@@ -2,18 +2,8 @@
 
 from Configuration.ProcessModifiers.enableSonicTriton_cff import enableSonicTriton
 
-_gpu_available_cached = None
-
-def _gpu_available():
-    global _gpu_available_cached
-    if _gpu_available_cached is None:
-        import os
-        _gpu_available_cached = (os.system("nvidia-smi -L") == 0)
-    return _gpu_available_cached
-
 enableSonicTriton.toModify(TritonService,
     fallback = dict(
         enable = True,
-        useGPU = _gpu_available(),
     ),
 )