Add multi-node job support to swm-core

HOWTO/JOB_DIRECTIVES.md

@@ -0,0 +1,153 @@
+# Job Script Directives
+
+Job scripts in Sky Port use special directives prefixed with `#SWM` to specify job requirements and configuration.
+
+## Available Directives
+
+### Resource Requirements
+
+#### nodes
+Specify the number of nodes to allocate for the job.
+```bash
+#SWM nodes <count>
+```
+**Example:**
+```bash
+#SWM nodes 3
+```
+This will request a partition with 3 compute nodes. The nodes will be named:
+- `swm-<jobid>-main` (primary/manager node)
+- `swm-<jobid>-node0` (first extra compute node)
+- `swm-<jobid>-node1` (second extra compute node)
+
+**Default:** 1 node
+
+**Note:** Multi-node job support requires the gate to support the multi-node partition feature (available in swm-cloud-gate branch `feature/multi-node-jobs`).
+
+#### flavor
+Specify the cloud instance flavor/size.
+```bash
+#SWM flavor <flavor_name>
+```
+
+#### gpus
+Request GPU resources.
+```bash
+#SWM gpus <count>
+```
+
+### Image Configuration
+
+#### cloud-image
+Specify the cloud VM image to use.
+```bash
+#SWM cloud-image <image_name>
+```
+
+#### container-image
+Specify the Docker container image to run the job in.
+```bash
+#SWM container-image <image:tag>
+```
+
+### Job Metadata
+
+#### name
+Set a human-readable name for the job.
+```bash
+#SWM name <job_name>
+```
+
+#### comment
+Add a description/comment for the job.
+```bash
+#SWM comment <description>
+```
+
+#### account
+Specify the account to use for billing.
+```bash
+#SWM account <account_name>
+```
+
+### Input/Output
+
+#### stdin
+Specify the standard input file.
+```bash
+#SWM stdin <file_path>
+```
+
+#### stdout
+Specify where to redirect standard output.
+```bash
+#SWM stdout <file_path>
+```
+
+#### stderr
+Specify where to redirect standard error.
+```bash
+#SWM stderr <file_path>
+```
+
+#### workdir
+Set the working directory for the job.
+```bash
+#SWM workdir <directory_path>
+```
+
+#### input-files
+Specify input files to be transferred to the job.
+```bash
+#SWM input-files <file1> <file2> ...
+```
+
+#### output-files
+Specify output files to be transferred back after job completion.
+```bash
+#SWM output-files <file1> <file2> ...
+```
+
+### Networking
+
+#### ports
+Specify ports to forward from the remote node.
+```bash
+#SWM ports <port1>,<port2>,...
+```
+
+#### submission-address
+Specify the submission address.
+```bash
+#SWM submission-address <address>
+```
+
+### Job Behavior
+
+#### relocatable
+Mark the job as relocatable (can be migrated between nodes).
+```bash
+#SWM relocatable
+```
+
+## Complete Multi-Node Example
+
+```bash
+#!/bin/bash
+
+#SWM name Multi-Node MPI Job
+#SWM nodes 4
+#SWM flavor Standard_D4s_v3
+#SWM cloud-image ubuntu22.04
+#SWM container-image mpi/ubuntu:latest
+#SWM relocatable
+#SWM comment Distributed MPI computation across 4 nodes
+
+echo "Starting multi-node job on $(hostname)"
+echo "Job ID: $SWM_JOBID"
+
+# Your MPI or distributed application code here
+mpirun -n 16 ./my_parallel_app
+
+echo "Job completed"
+```

priv/examples/jobscripts/multi-node.job

@@ -0,0 +1,21 @@
+#!/bin/bash
+
+#SWM name Multi-Node Job Example
+#SWM nodes 3
+#SWM relocatable
+#SWM comment Example job script demonstrating multi-node job submission
+
+# This job requests 3 nodes from the cloud provider
+# The nodes will be named: swm-<jobid>-node0, swm-<jobid>-node1, swm-<jobid>-node2
+# where swm-<jobid>-node0 is the main/master node
+
+echo "Hello from multi-node job $SWM_JOBID"
+echo "Running on node: $(hostname)"
+date
+
+# Your multi-node application code here
+# For example, MPI applications, distributed computing tasks, etc.
+
+sleep 60
+
+echo "Multi-node job completed"

src/srv/user/wm_jobscript.erl

@@ -102,6 +102,8 @@ parse_line(Ws, Job) when hd(Ws) == "flavor", length(Ws) > 1 ->
     add_requested_resource("flavor", lists:flatten(tl(Ws)), Job);
 parse_line(Ws, Job) when hd(Ws) == "gpus", length(Ws) > 1 ->
     add_requested_resource("gpus", list_to_integer(lists:flatten(tl(Ws))), [], Job);
+parse_line(Ws, Job) when hd(Ws) == "nodes", length(Ws) > 1 ->
+    add_requested_resource("node", list_to_integer(lists:flatten(tl(Ws))), [], Job);
 parse_line(Ws, Job) when hd(Ws) == "account", length(Ws) > 1 ->
     wm_entity:set({account_id, get_account_id(lists:flatten(tl(Ws)))}, Job);
 parse_line(Ws, Job) when hd(Ws) == "name", length(Ws) > 1 ->
@@ -135,3 +137,31 @@ parse_line(Ws, Job) ->
 get_account_id(AccountName) ->
     {ok, Account} = wm_conf:select(account, {name, AccountName}),
     wm_entity:get(id, Account).
+
+%% ============================================================================
+%% Tests
+%% ============================================================================
+
+-ifdef(EUNIT).
+
+-include_lib("eunit/include/eunit.hrl").
+
+% ./rebar3 eunit --module=wm_jobscript
+-spec parse_nodes_directive_test() -> ok.
+parse_nodes_directive_test() ->
+    JobScript = "#!/bin/bash\n#SWM nodes 3\necho hello",
+    Job = parse(JobScript),
+    Resources = wm_entity:get(request, Job),
+    NodeResource = lists:keyfind("node", 2, Resources),
+    ?assertMatch(#resource{name = "node", count = 3}, NodeResource).
+
+-spec parse_single_node_default_test() -> ok.
+parse_single_node_default_test() ->
+    JobScript = "#!/bin/bash\necho hello",
+    Job = parse(JobScript),
+    Resources = wm_entity:get(request, Job),
+    % When no nodes directive is present, default should be added elsewhere
+    % This test just ensures parsing doesn't crash
+    ?assertEqual([], Resources).
+
+-endif.