Attempt to use versioned container

Check if it's locally available, then try and pull. Then just use
latest if it's not available.

Signed-off-by: Eric Curtin <[email protected]>

ramalama/common.py

@@ -58,7 +58,7 @@ def available(cmd):
     return shutil.which(cmd) is not None
 
 
-def exec_cmd(args, stderr=True, debug=False):
+def exec_cmd(args, debug=False):
     if debug:
         perror("exec_cmd: ", *args)
 
@@ -69,22 +69,37 @@ def exec_cmd(args, stderr=True, debug=False):
         raise
 
 
-def run_cmd(args, cwd=None, stdout=subprocess.PIPE, ignore_stderr=False, debug=False):
+def run_cmd(args, cwd=None, stdout=subprocess.PIPE, ignore_stderr=False, ignore_all=False, debug=False):
     """
     Run the given command arguments.
 
     Args:
     args: command line arguments to execute in a subprocess
     cwd: optional working directory to run the command from
+    stdout: standard output configuration
+    ignore_stderr: if True, ignore standard error
+    ignore_all: if True, ignore both standard output and standard error
+    debug: if True, print debug information
     """
     if debug:
         perror("run_cmd: ", *args)
+        perror(f"Working directory: {cwd}")
+        perror(f"Ignore stderr: {ignore_stderr}")
+        perror(f"Ignore all: {ignore_all}")
 
-    stderr = None
-    if ignore_stderr:
-        stderr = subprocess.PIPE
+    serr = stderr
+    if ignore_all or ignore_stderr:
+        serr = subprocess.DEVNULL
 
-    return subprocess.run(args, check=True, cwd=cwd, stdout=stdout, stderr=stderr)
+    sout = stdout
+    if ignore_all:
+        sout = subprocess.DEVNULL
+
+    result = subprocess.run(args, check=True, cwd=cwd, stdout=sout, stderr=serr)
+    if debug:
+        print("Command finished with return code:", result.returncode)
+
+    return result
 
 
 def find_working_directory():
@@ -143,7 +158,8 @@ def default_image():
     image = os.getenv("RAMALAMA_IMAGE")
     if image:
         return image
-    return "quay.io/ramalama/ramalama:latest"
+
+    return "quay.io/ramalama/ramalama"
 
 
 def genname():

ramalama/model.py

@@ -4,6 +4,7 @@
 import shlex
 
 from ramalama.common import (
+    container_manager,
     default_image,
     exec_cmd,
     find_working_directory,
@@ -94,6 +95,18 @@ def remove(self, args):
                 raise KeyError(f"removing {self.model}: {e}")
         self.garbage_collection(args)
 
+    def attempt_to_use_versioned(self, conman, image, vers, args):
+        try:
+            if run_cmd([conman, "inspect", f"{image}:{vers}"], ignore_all=True, debug=args.debug):
+                return True
+
+            return run_cmd([conman, "pull", f"{image}:{vers}"], debug=args.debug)
+
+        except Exception:
+            return False
+
+        return False
+
     def _image(self, args):
         if args.image != default_image():
             return args.image
@@ -105,14 +118,19 @@ def _image(self, args):
 
             return "quay.io/modh/vllm:rhoai-2.17-cuda"
 
-        if gpu_type == "HIP_VISIBLE_DEVICES":
-            return "quay.io/ramalama/rocm:latest"
-        elif gpu_type == "CUDA_VISIBLE_DEVICES":
-            return "quay.io/ramalama/cuda:latest"
-        elif gpu_type == "ASAHI_VISIBLE_DEVICES":
-            return "quay.io/ramalama/asahi:latest"
+        vers = version()
+        conman = container_manager()
+        images = {
+            "HIP_VISIBLE_DEVICES": "quay.io/ramalama/rocm",
+            "CUDA_VISIBLE_DEVICES": "quay.io/ramalama/cuda",
+            "ASAHI_VISIBLE_DEVICES": "quay.io/ramalama/asahi",
+        }
+
+        image = images.get(gpu_type, args.image)
+        if self.attempt_to_use_versioned(conman, image, vers, args):
+            return f"{image}:{vers}"
 
-        return args.image
+        return f"{image}:latest"
 
     def setup_container(self, args):
         if hasattr(args, "name") and args.name:

test/system/015-help.bats

@@ -126,7 +126,7 @@ function check_help() {
 
     run_ramalama --help
     is "$output" ".*image IMAGE.*OCI container image to run with the specified AI model"  "Verify default image"
-    is "$output" ".*default: quay.io/ramalama/ramalama:latest"  "Verify default image"
+    is "$output" ".*default: quay.io/ramalama/ramalama"  "Verify default image"
 
     image=m_$(safename)
     RAMALAMA_IMAGE=${image} run_ramalama --help

test/system/030-run.bats

@@ -31,7 +31,7 @@ load helpers
 	is "${lines[0]}"  "Error: --nocontainer and --name options conflict. --name requires a container." "conflict between nocontainer and --name line"
 
 	RAMALAMA_IMAGE=${image} run_ramalama --dryrun run ${model}
-	is "$output" ".*${image} /bin/sh -c" "verify image name"
+	is "$output" ".*${image}:latest /bin/sh -c" "verify image name"
     else
 	run_ramalama --dryrun run -c 4096 ${model}
 	is "$output" 'llama-run -c 4096 --temp 0.8 /path/to/model.*' "dryrun correct"

test/system/040-serve.bats

@@ -222,7 +222,7 @@ verify_begin=".*run --rm -i --label RAMALAMA --security-opt=label=disable --name
 	is "$output" ".*command: \[\"vllm\"\]" "command is correct"
 	is "$output" ".*args: \['serve', '--port', '1234', '/mnt/models'\]" "args is correct"
 
-	is "$output" ".*image: quay.io/ramalama/ramalama:latest" "image is correct"
+	is "$output" ".*image: quay.io/ramalama/ramalama" "image is correct"
 	is "$output" ".*reference: ${ociimage}" "AI image should be created"
 	is "$output" ".*pullPolicy: IfNotPresent" "pullPolicy should exist"
 
@@ -240,7 +240,7 @@ verify_begin=".*run --rm -i --label RAMALAMA --security-opt=label=disable --name
     is "$output" ".*Generating Kubernetes YAML file: ${name}.yaml" "generate .yaml file"
 
     run cat $name.yaml
-    is "$output" ".*image: quay.io/ramalama/ramalama:latest" "Should container image"
+    is "$output" ".*image: quay.io/ramalama/ramalama" "Should container image"
     is "$output" ".*command: \[\"llama-server\"\]" "Should command"
     is "$output" ".*containerPort: 1234" "Should container container port"
 
@@ -257,7 +257,7 @@ verify_begin=".*run --rm -i --label RAMALAMA --security-opt=label=disable --name
     is "$output" ".*Generating quadlet file: ${name}.kube" "generate .kube file"
 
     run cat $name.yaml
-    is "$output" ".*image: quay.io/ramalama/ramalama:latest" "Should container image"
+    is "$output" ".*image: quay.io/ramalama/ramalama" "Should container image"
     is "$output" ".*command: \[\"llama-server\"\]" "Should command"
     is "$output" ".*containerPort: 1234" "Should container container port"
 

test/system/060-info.bats

@@ -20,7 +20,7 @@ load helpers
 
     # FIXME Engine  (podman|docker|'')
     tests="
-Image   | "quay.io/ramalama/ramalama:latest"
+Image   | "quay.io/ramalama/ramalama"
 Runtime | "llama.cpp"
 Version | "${version}"
 Store   | \\\("${HOME}/.local/share/ramalama"\\\|"/var/lib/ramalama"\\\)

test/system/helpers.podman.bash

@@ -300,7 +300,7 @@ function leak_check() {
 
     # FIXME Images. Exclude our standard expected images.
 #    _run_podman_quiet images --all --format '{{.ID}} {{.Repository}}:{{.Tag}}'
-#    output=$(awk "\$2 != \"$IMAGE\" && \$2 != \"quay.io/ramalama/ramalama:latest\" && \$2 !~ \"localhost/podman-pause:\" { print }" <<<"$output")
+#    output=$(awk "\$2 != \"$IMAGE\" && \$2 != \"quay.io/ramalama/ramalama\" && \$2 !~ \"localhost/podman-pause:\" { print }" <<<"$output")
 #    _leak_check_one "image"
 
     return $exit_code