Implement SSH user data handling and adjust volume size
- Added functions to generate SSH user data in cloud-config format and encode it in Base64 for server creation. - Updated the `create_gpu_server` function to include public key handling for SSH access. - Adjusted the default boot volume size from 40GB to 30GB in `os_client.py`. - Enhanced the `cmd_up` function to manage SSH key injection and server state more effectively. - Improved logging for SSH connection attempts and error handling in `wait_ssh` to provide clearer feedback on authentication issues.
This commit is contained in:
+57
-8
@@ -335,6 +335,30 @@ def _tag_server(conn, server, spot: bool) -> None:
|
||||
pass
|
||||
|
||||
|
||||
def _ssh_user_data(public_key: str) -> str:
|
||||
"""Minimal cloud-config (Selectel docs: ssh_authorized_keys)."""
|
||||
key = public_key.strip()
|
||||
return (
|
||||
"#cloud-config\n"
|
||||
"ssh_authorized_keys:\n"
|
||||
f" - {key}\n"
|
||||
"users:\n"
|
||||
" - default\n"
|
||||
" - name: ubuntu\n"
|
||||
" lock_passwd: true\n"
|
||||
" shell: /bin/bash\n"
|
||||
" sudo: ['ALL=(ALL) NOPASSWD:ALL']\n"
|
||||
" ssh_authorized_keys:\n"
|
||||
f" - {key}\n"
|
||||
)
|
||||
|
||||
|
||||
def _user_data_b64(public_key: str) -> str:
|
||||
import base64
|
||||
|
||||
return base64.b64encode(_ssh_user_data(public_key).encode("utf-8")).decode("ascii")
|
||||
|
||||
|
||||
def create_gpu_server(
|
||||
conn,
|
||||
*,
|
||||
@@ -345,6 +369,7 @@ def create_gpu_server(
|
||||
data_volume_id: str,
|
||||
az: str,
|
||||
spot: bool,
|
||||
public_key: str,
|
||||
log: Callable[[str], None],
|
||||
) -> Any:
|
||||
bdm = [
|
||||
@@ -366,7 +391,7 @@ def create_gpu_server(
|
||||
tags = [RESOURCE_TAG]
|
||||
if spot:
|
||||
tags.append(PREEMPTIBLE_TAG)
|
||||
kwargs: dict[str, Any] = {
|
||||
base: dict[str, Any] = {
|
||||
"name": SERVER_NAME,
|
||||
"flavor_id": flavor_id,
|
||||
"networks": [{"uuid": net_id}],
|
||||
@@ -374,16 +399,40 @@ def create_gpu_server(
|
||||
"availability_zone": az,
|
||||
"block_device_mapping_v2": bdm,
|
||||
"security_groups": [{"name": sg_name}],
|
||||
"tags": tags,
|
||||
}
|
||||
try:
|
||||
server = conn.compute.create_server(**kwargs)
|
||||
except Exception as exc:
|
||||
kwargs.pop("tags", None)
|
||||
ud_b64 = _user_data_b64(public_key)
|
||||
ud_plain = _ssh_user_data(public_key)
|
||||
|
||||
# Selectel: OpenStack API expects Base64 user_data; config_drive often unsupported.
|
||||
attempts: list[tuple[str, dict[str, Any]]] = [
|
||||
("tags+user_data_b64", {**base, "tags": tags, "user_data": ud_b64}),
|
||||
("user_data_b64", {**base, "user_data": ud_b64}),
|
||||
("user_data_plain", {**base, "user_data": ud_plain}),
|
||||
("key_name_only", {**base, "tags": tags}),
|
||||
("minimal", dict(base)),
|
||||
]
|
||||
server = None
|
||||
last_exc: BaseException | None = None
|
||||
used = ""
|
||||
for label, kwargs in attempts:
|
||||
try:
|
||||
server = conn.compute.create_server(**kwargs)
|
||||
except Exception as exc2:
|
||||
raise _wrap(exc2, "create server") from exc
|
||||
used = label
|
||||
break
|
||||
except Exception as exc:
|
||||
last_exc = exc
|
||||
log(f"create {label}: {exc}")
|
||||
if server is None:
|
||||
raise _wrap(last_exc or RuntimeError("create failed"), "create server")
|
||||
|
||||
if "user_data" not in used:
|
||||
log(
|
||||
f"create OK ({used}) без user_data — ключ может не попасть на boot-from-volume; "
|
||||
"смотри docs Selectel user-data (Base64)"
|
||||
)
|
||||
else:
|
||||
log(f"create OK ({used})")
|
||||
|
||||
log("ждём Nova ACTIVE (GPU create может занять несколько минут)")
|
||||
server = wait_server(conn, server, "ACTIVE")
|
||||
_tag_server(conn, server, spot)
|
||||
|
||||
Reference in New Issue
Block a user