tao.chen/mcp-server
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(yarn_client): use amContainerLogs for AM container log fallback

Browse Source 
The previous _fetch_logs_via_nodemanager walked
/apps/{appid}/appattempts -> /apps/{appid}/appattempts/{id}/containers,
but YARN ResourceManager does not expose the second hop. Container
info lives on NodeManager and is not reachable from RM REST, so the
fallback returned _logs_unavailable_error on every cluster without
/aggregated-logs (Hadoop 2.x, log aggregation disabled).

Replace the walk with: GET /apps/{appid} -> read app.amContainerLogs
-> GET {amContainerLogs}/stdout. The URL is provided directly by the
app response, so no container enumeration is needed.

Returns AM (driver) container logs only. Executor container logs
still require yarn.log-aggregation-enable=true (primary /aggregated-logs
path); the existing _logs_unavailable_error message already says that,
so the contract is preserved.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Claude
2026-06-30 10:33:01 +08:00
co-authored by Claude Fable 5
parent dfca612f8c
commit fd0036ca80
2 changed files with 44 additions and 93 deletions
Download Patch File Download Diff File Expand all files Collapse all files
spark_executor/core/yarn_client.py
+25 -42
View File
@@ -8,9 +8,15 @@ so the runtime image does not need a Hadoop client installation — the
`httpx` library already in pyproject.toml is enough.
Endpoints used (YARN 2.6+):
GET /ws/v1/cluster/apps/{appid} -> app status + state
GET /ws/v1/cluster/apps/{appid} -> app status + state; amContainerLogs
GET /ws/v1/cluster/apps/{appid}/aggregated-logs -> aggregated container logs
PUT /ws/v1/cluster/apps/{appid}/state -> kill an app (body: {"state":"KILLED"})
PUT /ws/v1/cluster/apps/{appid}/state -> kill an app (body: {"state":"KILLED"})
When aggregated logs are unavailable (HTTP 404/501, e.g. log aggregation
disabled), fall back to the amContainerLogs field reported by the app
endpoint and fetch the AM (driver) container's /stdout directly from the
NodeManager. This returns driver logs only; executor container logs still
require yarn.log-aggregation-enable=true (the primary /aggregated-logs path).
The ResourceManager URL is passed in per call (snapshotted on the Job at
confirm_submit_job time) and falls back to the YARN_RESOURCE_MANAGER_URL env
@@ -157,55 +163,32 @@ def _logs_unavailable_error(application_id: str) -> YarnError:
)
def _fetch_logs_via_nodemanager(application_id: str, config: YarnClientConfig) -> str:
def _fetch_logs_via_am_container(application_id: str, config: YarnClientConfig) -> str:
"""
Hadoop 2.x fallback: walk app attempts -> containers -> NodeManager
container logs and concatenate the results.
Fetch ApplicationMaster (driver) container stdout as a fallback when the
aggregated-logs endpoint is unavailable.
Returns AM (driver) container logs only. Executor container logs require
yarn.log-aggregation-enable=true, which is covered by the primary
/aggregated-logs path.
"""
base = _base_url(config.yarn_rm_url)
app_url = f"{base}/ws/v1/cluster/apps/{application_id}"
# Latest (or all) app attempts.
attempts_url = f"{app_url}/appattempts"
resp = _request("GET", attempts_url, timeout=30.0, verify=config.verify_for_httpx(), auth=config.auth_for_httpx())
resp = _request("GET", app_url, verify=config.verify_for_httpx(), auth=config.auth_for_httpx())
if resp.status_code >= 400:
raise _logs_unavailable_error(application_id)
attempts = resp.json().get("appAttempts", {}).get("appAttempt", [])
if not attempts:
am_container_logs = resp.json().get("app", {}).get("amContainerLogs")
if not am_container_logs:
raise _logs_unavailable_error(application_id)
# Collect containers across attempts.
containers: list[dict] = []
for attempt in attempts:
attempt_id = attempt.get("id")
if not attempt_id:
continue
containers_url = f"{app_url}/appattempts/{attempt_id}/containers"
resp = _request("GET", containers_url, timeout=30.0, verify=config.verify_for_httpx(), auth=config.auth_for_httpx())
if resp.status_code >= 400:
raise _logs_unavailable_error(application_id)
containers.extend(resp.json().get("containers", {}).get("container", []))
if not containers:
log_url = f"{am_container_logs.rstrip('/')}/stdout"
resp = _request("GET", log_url, timeout=60.0, verify=config.verify_for_httpx(), auth=config.auth_for_httpx())
if resp.status_code >= 400:
logger.error(f"YARN GET {log_url} -> {resp.status_code}: {resp.text[:500]}")
raise _logs_unavailable_error(application_id)
parts = []
for container in containers:
container_id = container.get("id")
user = container.get("user")
node_http_address = container.get("nodeHttpAddress")
if not all((container_id, user, node_http_address)):
continue
scheme = "https" if config.yarn_rm_url and config.yarn_rm_url.startswith("https://") else "http"
nm_url = f"{scheme}://{node_http_address}/node/containerlogs/{container_id}/{user}/"
resp = _request("GET", nm_url, timeout=60.0, verify=config.verify_for_httpx(), auth=config.auth_for_httpx())
if resp.status_code >= 400:
raise _logs_unavailable_error(application_id)
parts.append(f"=== container: {container_id} ===\n{resp.text}")
if not parts:
raise _logs_unavailable_error(application_id)
return "\n\n".join(parts)
return resp.text
def get_application_logs(application_id: str, config: YarnClientConfig) -> str:
@@ -213,7 +196,7 @@ def get_application_logs(application_id: str, config: YarnClientConfig) -> str:
url = f"{_base_url(config.yarn_rm_url)}/ws/v1/cluster/apps/{application_id}/aggregated-logs"
resp = _request("GET", url, timeout=60.0, verify=config.verify_for_httpx(), auth=config.auth_for_httpx())
if resp.status_code in (404, 501):
return _fetch_logs_via_nodemanager(application_id, config)
return _fetch_logs_via_am_container(application_id, config)
if resp.status_code >= 400:
logger.error(f"YARN GET {url} -> {resp.status_code}: {resp.text[:500]}")
raise YarnError(f"YARN GET logs returned HTTP {resp.status_code}")
tests/unit/test_yarn_client.py
+19 -51
View File
@@ -125,52 +125,40 @@ def test_logs_returns_text_on_h3_aggregated_endpoint():
assert m.call_args.kwargs["verify"] is True
def test_logs_falls_back_to_nodemanager_on_404():
def test_logs_falls_back_to_am_container_on_404():
responses = [
_resp(404), # aggregated-logs H3 endpoint missing
_resp(200, json_data={"appAttempts": {"appAttempt": [{"id": "attempt_1"}]}}),
_resp(404), # aggregated-logs endpoint missing
_resp(
200,
json_data={
"containers": {
"container": [
{
"id": "container_1",
"user": "hdfs",
"nodeHttpAddress": "nm1:8042",
}
]
"app": {
"amContainerLogs": "http://nm1:8042/node/containerlogs/container_1/hdfs",
}
},
),
_resp(200, text="container log content"),
_resp(200, text="am container log content"),
]
with patch(
"spark_executor.core.yarn_client.httpx.request", side_effect=responses
) as m:
out = get_application_logs("application_1", YarnClientConfig(yarn_rm_url=RM))
assert "container_1" in out
assert "container log content" in out
assert out == "am container log content"
calls = m.call_args_list
assert calls[0].args == (
"GET",
f"{RM}/ws/v1/cluster/apps/application_1/aggregated-logs",
)
assert calls[1].args == ("GET", f"{RM}/ws/v1/cluster/apps/application_1/appattempts")
assert calls[1].args == ("GET", f"{RM}/ws/v1/cluster/apps/application_1")
assert calls[2].args == (
"GET",
f"{RM}/ws/v1/cluster/apps/application_1/appattempts/attempt_1/containers",
)
assert calls[3].args == (
"GET",
"http://nm1:8042/node/containerlogs/container_1/hdfs/",
"http://nm1:8042/node/containerlogs/container_1/hdfs/stdout",
)
def test_logs_raises_on_404_when_nm_also_empty():
def test_logs_raises_on_404_when_am_container_field_missing():
responses = [
_resp(404), # aggregated-logs H3 endpoint missing
_resp(200, json_data={"appAttempts": {"appAttempt": []}}),
_resp(404), # aggregated-logs endpoint missing
_resp(200, json_data={"app": {}}),
]
with patch(
"spark_executor.core.yarn_client.httpx.request", side_effect=responses
@@ -179,47 +167,27 @@ def test_logs_raises_on_404_when_nm_also_empty():
get_application_logs("application_1", YarnClientConfig(yarn_rm_url=RM))
def test_logs_handles_multiple_containers():
def test_logs_returns_am_container_stdout():
responses = [
_resp(404), # aggregated-logs H3 endpoint missing
_resp(200, json_data={"appAttempts": {"appAttempt": [{"id": "attempt_1"}]}}),
_resp(404), # aggregated-logs endpoint missing
_resp(
200,
json_data={
"containers": {
"container": [
{
"id": "container_1",
"user": "hdfs",
"nodeHttpAddress": "nm1:8042",
},
{
"id": "container_2",
"user": "hdfs",
"nodeHttpAddress": "nm2:8042",
},
]
"app": {
"amContainerLogs": "http://nm1:8042/node/containerlogs/container_1/hdfs",
}
},
),
_resp(200, text="log one"),
_resp(200, text="log two"),
_resp(200, text="driver stdout"),
]
with patch(
"spark_executor.core.yarn_client.httpx.request", side_effect=responses
) as m:
out = get_application_logs("application_1", YarnClientConfig(yarn_rm_url=RM))
assert "log one" in out
assert "log two" in out
assert out.index("log one") < out.index("log two")
calls = m.call_args_list
assert calls[3].args == (
assert out == "driver stdout"
assert m.call_args_list[2].args == (
"GET",
"http://nm1:8042/node/containerlogs/container_1/hdfs/",
)
assert calls[4].args == (
"GET",
"http://nm2:8042/node/containerlogs/container_2/hdfs/",
"http://nm1:8042/node/containerlogs/container_1/hdfs/stdout",
)