feat(cli): add beta clusters recreate command

src/together/lib/cli/__init__.py

@@ -3,7 +3,7 @@
 import os
 import sys
 import inspect
-from typing import Optional, Annotated, get_args, get_origin
+from typing import Literal, Optional, Annotated, get_args, get_origin
 
 import httpx
 from cyclopts import App, Group, Parameter, CycloptsError, MissingArgumentError
@@ -112,6 +112,13 @@
 # stripped; reported separately via is_beta_command).
 _NO_AUTH_COMMANDS = frozenset({"clusters ssh"})
 
+# Internal environment shorthands for the hidden --env flag. An explicit
+# --base-url wins; TOGETHER_BASE_URL only applies when neither flag is given.
+_ENV_BASE_URLS = {
+    "qa": "https://api.qa.together.ai/v1",
+    "prod": "https://api.together.ai/v1",
+}
+
 
 async def _resolve_project_id(client: AsyncTogether) -> str:
     me = await client.whoami()
@@ -205,6 +212,7 @@ async def launcher(
     *tokens: Annotated[str, Parameter(show=False, allow_leading_hyphen=True)],
     api_key: Annotated[Optional[str], Parameter(show=False)] = None,
     base_url: Annotated[Optional[str], Parameter(show=False)] = None,
+    env: Annotated[Optional[Literal["qa", "prod"]], Parameter(show=False)] = None,
     timeout: Annotated[Optional[int], Parameter(show=False)] = None,
     max_retries: Annotated[Optional[int], Parameter(show=False)] = None,
     debug: Annotated[Optional[bool], Parameter(show=False)] = False,
@@ -244,6 +252,9 @@ async def launcher(
     # they stay keyless.
     no_auth_command = is_beta_command and parsed_command in _NO_AUTH_COMMANDS
 
+    if base_url is None and env is not None:
+        base_url = _ENV_BASE_URLS[env]
+
     client = _create_client(api_key, base_url, timeout, max_retries, project_id, require_api_key=not no_auth_command)
 
     # Skip the project-resolution whoami() for out-of-band-auth commands: it is a
@@ -564,6 +575,10 @@ async def run_command() -> None:
     help_epilogue=BETA_CLUSTERS_UPDATE_HELP_EXAMPLES,
 )
 clusters_app.command((f"{_CLI}.beta.clusters.delete:delete"), alias="-d", help="Delete a cluster")
+clusters_app.command(
+    (f"{_CLI}.beta.clusters.recreate:recreate"),
+    help="Recreate a cluster, preserving its reservation",
+)
 clusters_app.command((f"{_CLI}.beta.clusters.list_regions:list_regions"), help="List regions for deploying clusters")
 clusters_app.command(
     (f"{_CLI}.beta.clusters.get_credentials:get_credentials"),

src/together/lib/cli/api/beta/clusters/recreate.py

@@ -0,0 +1,69 @@
+from __future__ import annotations
+
+from typing import Literal
+
+import httpx
+
+from together.lib.cli.utils.config import CLIConfigParameter
+from together.lib.cli.utils._console import console
+from together.lib.cli.components.loader import show_loading_status
+from together.lib.cli.api.beta.clusters._util import print_clusters
+
+
+async def recreate(
+    cluster_id: str,
+    *,
+    reason: str | None = None,
+    cluster_type: Literal["KUBERNETES", "SLURM"] | None = None,
+    num_gpus: int | None = None,
+    num_reserved_gpus: int | None = None,
+    num_capacity_pool_gpus: int | None = None,
+    num_preemptible_gpus: int | None = None,
+    config: CLIConfigParameter,
+) -> None:
+    """Tear down and rebuild a cluster while preserving its reservation.
+
+    Spec options (cluster type and GPU count targets) are applied as the
+    cluster's new spec; when none are given, the current spec is kept.
+    """
+
+    new_spec = {
+        key: value
+        for key, value in {
+            "cluster_type": cluster_type,
+            "num_gpus": num_gpus,
+            "num_reserved_gpus": num_reserved_gpus,
+            "num_capacity_pool_gpus": num_capacity_pool_gpus,
+            "num_preemptible_gpus": num_preemptible_gpus,
+        }.items()
+        if value is not None
+    }
+
+    body: dict[str, object] = {}
+    if reason:
+        body["reason"] = reason
+    if new_spec:
+        body["new_spec"] = new_spec
+
+    if config.json:
+        response = await config.client.post(
+            f"/compute/clusters/{cluster_id}:recreate",
+            cast_to=httpx.Response,
+            body=body,
+        )
+        console.print_json(response.text)
+        return
+
+    cluster = await show_loading_status("", config.client.beta.clusters.retrieve(cluster_id=cluster_id))
+    print_clusters([cluster])
+    resp = input(f"Clusters: Are you sure you want to recreate cluster {cluster.cluster_name}? [y/N] ").strip().lower()
+    if resp != "y" and resp != "yes":
+        return
+    response = await show_loading_status(
+        "Recreating cluster...",
+        config.client.post(f"/compute/clusters/{cluster_id}:recreate", cast_to=httpx.Response, body=body),
+    )
+    intent = response.json()
+    console.print(
+        f"Recreate requested for {cluster.cluster_name} ({cluster_id}); intent {intent.get('id')} is {intent.get('status')}"
+    )

tests/cli/test_beta_clusters.py

@@ -1153,6 +1153,60 @@ def test_delete_confirm_yes(self, respx_mock: MockRouter, cli_runner: CliRunner)
         assert result.exit_code == 0
 
 
+class TestBetaClustersRecreate:
+    @pytest.mark.respx(base_url=base_url)
+    def test_recreate_json(self, respx_mock: MockRouter, cli_runner: CliRunner) -> None:
+        respx_mock.post("/compute/clusters/c-rec:recreate").mock(
+            return_value=httpx.Response(200, json={"id": "intent-1", "cluster_id": "c-rec", "status": "pending"})
+        )
+        result = cli_runner.invoke(["beta", "clusters", "recreate", "c-rec", "--json"])
+        assert json.loads(result.output) == {"id": "intent-1", "cluster_id": "c-rec", "status": "pending"}
+        assert result.exit_code == 0
+
+    @pytest.mark.respx(base_url=base_url)
+    def test_recreate_confirm_yes_sends_reason(self, respx_mock: MockRouter, cli_runner: CliRunner) -> None:
+        c = _cluster_body("c1", "to-recreate")
+        respx_mock.get("/compute/clusters/c1").mock(return_value=httpx.Response(200, json=c))
+        route = respx_mock.post("/compute/clusters/c1:recreate").mock(
+            return_value=httpx.Response(200, json={"id": "intent-1", "cluster_id": "c1", "status": "pending"})
+        )
+        result = cli_runner.invoke(["beta", "clusters", "recreate", "c1", "--reason", "maintenance"], input="y\n")
+        assert "Recreate requested" in result.output
+        assert "intent-1" in result.output
+        assert result.exit_code == 0
+        sent = json.loads(route.calls.last.request.content)
+        assert sent == {"reason": "maintenance"}
+
+    @pytest.mark.respx(base_url=base_url)
+    def test_recreate_with_new_spec(self, respx_mock: MockRouter, cli_runner: CliRunner) -> None:
+        route = respx_mock.post("/compute/clusters/c1:recreate").mock(
+            return_value=httpx.Response(200, json={"id": "intent-2", "cluster_id": "c1", "status": "pending"})
+        )
+        result = cli_runner.invoke(
+            ["beta", "clusters", "recreate", "c1", "--cluster-type", "SLURM", "--num-gpus", "16", "--json"]
+        )
+        assert result.exit_code == 0
+        sent = json.loads(route.calls.last.request.content)
+        assert sent == {"new_spec": {"cluster_type": "SLURM", "num_gpus": 16}}
+
+    @pytest.mark.respx(base_url="https://api.qa.together.ai/v1")
+    def test_recreate_env_flag_targets_qa(self, respx_mock: MockRouter, cli_runner: CliRunner) -> None:
+        respx_mock.post("/compute/clusters/c-qa:recreate").mock(
+            return_value=httpx.Response(200, json={"id": "intent-3", "cluster_id": "c-qa", "status": "pending"})
+        )
+        result = cli_runner.invoke(["--env", "qa", "beta", "clusters", "recreate", "c-qa", "--json"])
+        assert result.exit_code == 0
+        assert json.loads(result.output)["id"] == "intent-3"
+
+    @pytest.mark.respx(base_url=base_url)
+    def test_recreate_confirm_no(self, respx_mock: MockRouter, cli_runner: CliRunner) -> None:
+        # No POST route is mocked: an unexpected recreate call would fail loudly.
+        c = _cluster_body("c1", "keep-me")
+        respx_mock.get("/compute/clusters/c1").mock(return_value=httpx.Response(200, json=c))
+        result = cli_runner.invoke(["beta", "clusters", "recreate", "c1"], input="n\n")
+        assert result.exit_code == 0
+
+
 class TestBetaClustersGetCredentials:
     @pytest.mark.respx(base_url=base_url)
     def test_get_credentials_stdout(self, respx_mock: MockRouter, cli_runner: CliRunner) -> None: