* misc backend fixes: - fix running w/o local minio - ensure crawler image pull policy is configurable, loaded via chart value - use digitalocean repo for main backend image (for now) - add bucket_name to config only if using default bucket * enable all behaviors, support 'access_endpoint_url' for default storages * debugging: add 'no_delete_jobs' setting for k8s and docker to disable deletion of completed jobs
708 lines
24 KiB
Python
708 lines
24 KiB
Python
""" K8s support"""
|
|
|
|
import os
|
|
import datetime
|
|
import json
|
|
import asyncio
|
|
import base64
|
|
|
|
from kubernetes_asyncio import client, config, watch
|
|
from kubernetes_asyncio.stream import WsApiClient
|
|
|
|
from crawls import Crawl, CrawlFile
|
|
|
|
|
|
# ============================================================================
|
|
CRAWLER_NAMESPACE = os.environ.get("CRAWLER_NAMESPACE") or "crawlers"
|
|
|
|
# an 2/31 schedule that will never run as empty is not allowed
|
|
DEFAULT_NO_SCHEDULE = "* * 31 2 *"
|
|
|
|
|
|
# ============================================================================
|
|
class K8SManager:
|
|
# pylint: disable=too-many-instance-attributes,too-many-locals,too-many-arguments
|
|
"""K8SManager, manager creation of k8s resources from crawl api requests"""
|
|
|
|
def __init__(self, namespace=CRAWLER_NAMESPACE):
|
|
config.load_incluster_config()
|
|
|
|
self.crawl_ops = None
|
|
|
|
self.core_api = client.CoreV1Api()
|
|
self.core_api_ws = client.CoreV1Api(api_client=WsApiClient())
|
|
self.batch_api = client.BatchV1Api()
|
|
self.batch_beta_api = client.BatchV1beta1Api()
|
|
|
|
self.namespace = namespace
|
|
self._default_storage_endpoints = {}
|
|
|
|
self.crawler_image = os.environ["CRAWLER_IMAGE"]
|
|
self.crawler_image_pull_policy = os.environ["CRAWLER_PULL_POLICY"]
|
|
|
|
self.crawl_retries = int(os.environ.get("CRAWL_RETRIES", "3"))
|
|
|
|
self.no_delete_jobs = os.environ.get("NO_DELETE_JOBS", "0") != "0"
|
|
|
|
self.loop = asyncio.get_running_loop()
|
|
self.loop.create_task(self.run_event_loop())
|
|
|
|
def set_crawl_ops(self, ops):
|
|
""" Set crawl ops handler """
|
|
self.crawl_ops = ops
|
|
|
|
async def run_event_loop(self):
|
|
""" Run the job watch loop, retry in case of failure"""
|
|
while True:
|
|
try:
|
|
await self.watch_events()
|
|
# pylint: disable=broad-except
|
|
except Exception as exc:
|
|
print(f"Retrying job loop: {exc}")
|
|
await asyncio.sleep(10)
|
|
|
|
async def watch_events(self):
|
|
""" Get events for completed jobs"""
|
|
async with watch.Watch().stream(
|
|
self.core_api.list_namespaced_event,
|
|
self.namespace,
|
|
field_selector="involvedObject.kind=Job",
|
|
) as stream:
|
|
async for event in stream:
|
|
try:
|
|
obj = event["object"]
|
|
if obj.reason == "BackoffLimitExceeded":
|
|
self.loop.create_task(
|
|
self.handle_crawl_failed(obj.involved_object.name, "failed")
|
|
)
|
|
|
|
elif obj.reason == "DeadlineExceeded":
|
|
self.loop.create_task(
|
|
self.handle_crawl_failed(
|
|
obj.involved_object.name, "timed_out"
|
|
)
|
|
)
|
|
|
|
# pylint: disable=broad-except
|
|
except Exception as exc:
|
|
print(exc)
|
|
|
|
# pylint: disable=unused-argument
|
|
async def get_storage(self, storage_name, is_default=False):
|
|
"""Check if storage_name is valid by checking existing secret
|
|
is_default flag ignored"""
|
|
try:
|
|
return await self.core_api.read_namespaced_secret(
|
|
f"storage-{storage_name}",
|
|
namespace=self.namespace,
|
|
)
|
|
except Exception:
|
|
# pylint: disable=broad-except,raise-missing-from
|
|
raise Exception(f"Storage {storage_name} not found")
|
|
|
|
return None
|
|
|
|
async def update_archive_storage(self, aid, uid, storage):
|
|
"""Update storage by either creating a per-archive secret, if using custom storage
|
|
or deleting per-archive secret, if using default storage"""
|
|
archive_storage_name = f"storage-{aid}"
|
|
if storage.type == "default":
|
|
try:
|
|
await self.core_api.delete_namespaced_secret(
|
|
archive_storage_name,
|
|
namespace=self.namespace,
|
|
propagation_policy="Foreground",
|
|
)
|
|
# pylint: disable=bare-except
|
|
except:
|
|
pass
|
|
|
|
return
|
|
|
|
labels = {"btrix.archive": aid, "btrix.user": uid}
|
|
|
|
crawl_secret = client.V1Secret(
|
|
metadata={
|
|
"name": archive_storage_name,
|
|
"namespace": self.namespace,
|
|
"labels": labels,
|
|
},
|
|
string_data={
|
|
"STORE_ENDPOINT_URL": storage.endpoint_url,
|
|
"STORE_ACCESS_KEY": storage.access_key,
|
|
"STORE_SECRET_KEY": storage.secret_key,
|
|
},
|
|
)
|
|
|
|
try:
|
|
await self.core_api.create_namespaced_secret(
|
|
namespace=self.namespace, body=crawl_secret
|
|
)
|
|
|
|
# pylint: disable=bare-except
|
|
except:
|
|
await self.core_api.patch_namespaced_secret(
|
|
name=archive_storage_name, namespace=self.namespace, body=crawl_secret
|
|
)
|
|
|
|
async def add_crawl_config(self, crawlconfig, storage, run_now):
|
|
"""add new crawl as cron job, store crawl config in configmap"""
|
|
cid = str(crawlconfig.id)
|
|
userid = crawlconfig.user
|
|
aid = crawlconfig.archive
|
|
|
|
annotations = {
|
|
"btrix.run.schedule": crawlconfig.schedule,
|
|
"btrix.storage_name": storage.name,
|
|
"btrix.colls": json.dumps(crawlconfig.colls),
|
|
}
|
|
|
|
# Configure Annotations + Labels
|
|
if storage.type == "default":
|
|
storage_name = storage.name
|
|
storage_path = storage.path
|
|
annotations["btrix.def_storage_path"] = storage_path
|
|
else:
|
|
storage_name = aid
|
|
storage_path = ""
|
|
|
|
labels = {
|
|
"btrix.user": userid,
|
|
"btrix.archive": aid,
|
|
"btrix.crawlconfig": cid,
|
|
}
|
|
|
|
await self.get_storage(storage_name)
|
|
|
|
# Create Config Map
|
|
config_map = self._create_config_map(crawlconfig, labels)
|
|
|
|
# Create Cron Job
|
|
await self.core_api.create_namespaced_config_map(
|
|
namespace=self.namespace, body=config_map
|
|
)
|
|
|
|
suspend, schedule = self._get_schedule_suspend_run_now(crawlconfig)
|
|
|
|
job_template = self._get_job_template(
|
|
cid,
|
|
storage_name,
|
|
storage_path,
|
|
labels,
|
|
annotations,
|
|
crawlconfig.crawlTimeout,
|
|
crawlconfig.parallel,
|
|
)
|
|
|
|
spec = client.V1beta1CronJobSpec(
|
|
schedule=schedule,
|
|
suspend=suspend,
|
|
concurrency_policy="Forbid",
|
|
successful_jobs_history_limit=2,
|
|
failed_jobs_history_limit=3,
|
|
job_template=job_template,
|
|
)
|
|
|
|
cron_job = client.V1beta1CronJob(
|
|
metadata={
|
|
"name": f"crawl-scheduled-{cid}",
|
|
"namespace": self.namespace,
|
|
"labels": labels,
|
|
},
|
|
spec=spec,
|
|
)
|
|
|
|
cron_job = await self.batch_beta_api.create_namespaced_cron_job(
|
|
namespace=self.namespace, body=cron_job
|
|
)
|
|
|
|
# Run Job Now
|
|
if run_now:
|
|
new_job = await self._create_run_now_job(cron_job)
|
|
return new_job.metadata.name
|
|
|
|
return ""
|
|
|
|
async def update_crawl_schedule(self, cid, schedule):
|
|
""" Update the schedule for existing crawl config """
|
|
|
|
cron_jobs = await self.batch_beta_api.list_namespaced_cron_job(
|
|
namespace=self.namespace, label_selector=f"btrix.crawlconfig={cid}"
|
|
)
|
|
|
|
if len(cron_jobs.items) != 1:
|
|
return
|
|
|
|
cron_job = cron_jobs.items[0]
|
|
|
|
real_schedule = schedule or DEFAULT_NO_SCHEDULE
|
|
|
|
if real_schedule != cron_job.spec.schedule:
|
|
cron_job.spec.schedule = real_schedule
|
|
cron_job.spec.suspend = not schedule
|
|
|
|
cron_job.spec.job_template.metadata.annotations[
|
|
"btrix.run.schedule"
|
|
] = schedule
|
|
|
|
await self.batch_beta_api.patch_namespaced_cron_job(
|
|
name=cron_job.metadata.name, namespace=self.namespace, body=cron_job
|
|
)
|
|
|
|
async def run_crawl_config(self, cid):
|
|
""" Run crawl job for cron job based on specified crawlconfig id (cid) """
|
|
cron_jobs = await self.batch_beta_api.list_namespaced_cron_job(
|
|
namespace=self.namespace, label_selector=f"btrix.crawlconfig={cid}"
|
|
)
|
|
|
|
if len(cron_jobs.items) != 1:
|
|
raise Exception("Crawl Config Not Found")
|
|
|
|
res = await self._create_run_now_job(cron_jobs.items[0])
|
|
return res.metadata.name
|
|
|
|
async def list_running_crawls(self, cid=None, aid=None, userid=None):
|
|
""" Return a list of running crawls """
|
|
filters = []
|
|
if cid:
|
|
filters.append(f"btrix.crawlconfig={cid}")
|
|
|
|
if aid:
|
|
filters.append(f"btrix.archive={aid}")
|
|
|
|
if userid:
|
|
filters.append(f"btrix.user={userid}")
|
|
|
|
jobs = await self.batch_api.list_namespaced_job(
|
|
namespace=self.namespace,
|
|
label_selector=",".join(filters),
|
|
field_selector="status.successful=0",
|
|
)
|
|
|
|
return [
|
|
self._make_crawl_for_job(job, "running")
|
|
for job in jobs.items
|
|
if job.status.active
|
|
]
|
|
|
|
async def init_crawl_screencast(self, crawl_id, aid):
|
|
""" Init service for this job/crawl_id to support screencasting """
|
|
labels = {"btrix.archive": aid}
|
|
|
|
service = client.V1Service(
|
|
kind="Service",
|
|
api_version="v1",
|
|
metadata={
|
|
"name": crawl_id,
|
|
"labels": labels,
|
|
},
|
|
spec={
|
|
"selector": {"job-name": crawl_id},
|
|
"ports": [{"protocol": "TCP", "port": 9037, "name": "screencast"}],
|
|
},
|
|
)
|
|
|
|
try:
|
|
await self.core_api.create_namespaced_service(
|
|
body=service, namespace=self.namespace
|
|
)
|
|
except client.exceptions.ApiException as api_exc:
|
|
if api_exc.status != 409:
|
|
raise api_exc
|
|
|
|
async def process_crawl_complete(self, crawlcomplete):
|
|
"""Ensure the crawlcomplete data is valid (job exists and user matches)
|
|
Fill in additional details about the crawl"""
|
|
job = await self.batch_api.read_namespaced_job(
|
|
name=crawlcomplete.id, namespace=self.namespace
|
|
)
|
|
|
|
if not job: # or job.metadata.labels["btrix.user"] != crawlcomplete.user:
|
|
return None, None
|
|
|
|
manual = job.metadata.annotations.get("btrix.run.manual") == "1"
|
|
if manual and not self.no_delete_jobs:
|
|
self.loop.create_task(self._delete_job(job.metadata.name))
|
|
|
|
crawl = self._make_crawl_for_job(
|
|
job,
|
|
"complete" if crawlcomplete.completed else "partial_complete",
|
|
finish_now=True,
|
|
)
|
|
|
|
storage_path = job.metadata.annotations.get("btrix.def_storage_path")
|
|
inx = None
|
|
filename = None
|
|
storage_name = None
|
|
if storage_path:
|
|
inx = crawlcomplete.filename.index(storage_path)
|
|
filename = (
|
|
crawlcomplete.filename[inx:] if inx > 0 else crawlcomplete.filename
|
|
)
|
|
storage_name = job.metadata.annotations.get("btrix.storage_name")
|
|
|
|
def_storage_name = storage_name if inx else None
|
|
|
|
crawl_file = CrawlFile(
|
|
def_storage_name=def_storage_name,
|
|
filename=filename or crawlcomplete.filename,
|
|
size=crawlcomplete.size,
|
|
hash=crawlcomplete.hash,
|
|
)
|
|
|
|
return crawl, crawl_file
|
|
|
|
async def get_default_storage_access_endpoint(self, name):
|
|
""" Get access_endpoint for default storage """
|
|
if name not in self._default_storage_endpoints:
|
|
storage_secret = await self.get_storage(name, is_default=True)
|
|
self._default_storage_endpoints[name] = base64.standard_b64decode(
|
|
storage_secret.data["STORE_ACCESS_ENDPOINT_URL"]
|
|
).decode()
|
|
|
|
return self._default_storage_endpoints[name]
|
|
|
|
async def is_running(self, job_name, aid):
|
|
""" Return true if the specified crawl (by job_name) is running """
|
|
try:
|
|
job = await self.batch_api.read_namespaced_job(
|
|
name=job_name, namespace=self.namespace
|
|
)
|
|
|
|
if not job or job.metadata.labels["btrix.archive"] != aid:
|
|
return False
|
|
|
|
return True
|
|
# pylint: disable=broad-except
|
|
except Exception:
|
|
return False
|
|
|
|
async def stop_crawl(self, job_name, aid, graceful=True):
|
|
"""Attempt to stop crawl, either gracefully by issuing a SIGTERM which
|
|
will attempt to finish current pages
|
|
|
|
OR, abruptly by first issueing a SIGINT, followed by SIGTERM, which
|
|
will terminate immediately"""
|
|
|
|
job = await self.batch_api.read_namespaced_job(
|
|
name=job_name, namespace=self.namespace
|
|
)
|
|
|
|
if not job or job.metadata.labels["btrix.archive"] != aid:
|
|
return None
|
|
|
|
result = None
|
|
|
|
if not graceful:
|
|
pods = await self.core_api.list_namespaced_pod(
|
|
namespace=self.namespace,
|
|
label_selector=f"job-name={job_name},btrix.archive={aid}",
|
|
)
|
|
|
|
await self._send_sig_to_pods(pods.items, aid)
|
|
|
|
result = self._make_crawl_for_job(job, "canceled", True)
|
|
else:
|
|
result = True
|
|
|
|
await self._delete_job(job_name)
|
|
|
|
return result
|
|
|
|
async def scale_crawl(self, job_name, aid, parallelism=1):
|
|
""" Set the crawl scale (job parallelism) on the specified job """
|
|
|
|
try:
|
|
job = await self.batch_api.read_namespaced_job(
|
|
name=job_name, namespace=self.namespace
|
|
)
|
|
# pylint: disable=broad-except
|
|
except Exception:
|
|
return "Crawl not found"
|
|
|
|
if not job or job.metadata.labels["btrix.archive"] != aid:
|
|
return "Invalid Crawled"
|
|
|
|
if parallelism < 1 or parallelism > 10:
|
|
return "Invalid Scale: Must be between 1 and 10"
|
|
|
|
job.spec.parallelism = parallelism
|
|
|
|
await self.batch_api.patch_namespaced_job(
|
|
name=job.metadata.name, namespace=self.namespace, body=job
|
|
)
|
|
|
|
return None
|
|
|
|
async def delete_crawl_configs_for_archive(self, archive):
|
|
"""Delete all crawl configs for given archive"""
|
|
return await self._delete_crawl_configs(f"btrix.archive={archive}")
|
|
|
|
async def delete_crawl_config_by_id(self, cid):
|
|
"""Delete all crawl configs by id"""
|
|
return await self._delete_crawl_configs(f"btrix.crawlconfig={cid}")
|
|
|
|
async def handle_crawl_failed(self, job_name, reason):
|
|
""" Handle failed crawl job, add to db and then delete """
|
|
try:
|
|
job = await self.batch_api.read_namespaced_job(
|
|
name=job_name, namespace=self.namespace
|
|
)
|
|
# pylint: disable=bare-except
|
|
except:
|
|
print("Job Failure Already Handled")
|
|
return
|
|
|
|
crawl = self._make_crawl_for_job(job, reason, True)
|
|
|
|
# if update succeeds, than crawl has not completed, so likely a failure
|
|
failure = await self.crawl_ops.store_crawl(crawl)
|
|
|
|
# keep failed jobs around, for now
|
|
if not failure and not self.no_delete_jobs:
|
|
await self._delete_job(job_name)
|
|
|
|
# ========================================================================
|
|
# Internal Methods
|
|
|
|
# pylint: disable=no-self-use
|
|
def _make_crawl_for_job(self, job, state, finish_now=False):
|
|
""" Make a crawl object from a job"""
|
|
return Crawl(
|
|
id=job.metadata.name,
|
|
state=state,
|
|
scale=job.spec.parallelism or 1,
|
|
user=job.metadata.labels["btrix.user"],
|
|
aid=job.metadata.labels["btrix.archive"],
|
|
cid=job.metadata.labels["btrix.crawlconfig"],
|
|
schedule=job.metadata.annotations.get("btrix.run.schedule", ""),
|
|
manual=job.metadata.annotations.get("btrix.run.manual") == "1",
|
|
started=job.status.start_time.replace(tzinfo=None),
|
|
finished=datetime.datetime.utcnow().replace(microsecond=0, tzinfo=None)
|
|
if finish_now
|
|
else None,
|
|
colls=json.loads(job.metadata.annotations.get("btrix.colls", [])),
|
|
)
|
|
|
|
async def _delete_job(self, name):
|
|
await self.batch_api.delete_namespaced_job(
|
|
name=name,
|
|
namespace=self.namespace,
|
|
grace_period_seconds=60,
|
|
propagation_policy="Foreground",
|
|
)
|
|
|
|
try:
|
|
await self.core_api.delete_namespaced_service(
|
|
name=name,
|
|
namespace=self.namespace,
|
|
grace_period_seconds=60,
|
|
propagation_policy="Foreground",
|
|
)
|
|
# pylint: disable=bare-except
|
|
except:
|
|
pass
|
|
|
|
def _create_config_map(self, crawlconfig, labels):
|
|
""" Create Config Map based on CrawlConfig + labels """
|
|
config_map = client.V1ConfigMap(
|
|
metadata={
|
|
"name": f"crawl-config-{crawlconfig.id}",
|
|
"namespace": self.namespace,
|
|
"labels": labels,
|
|
},
|
|
data={"crawl-config.json": json.dumps(crawlconfig.config.dict())},
|
|
)
|
|
|
|
return config_map
|
|
|
|
# pylint: disable=no-self-use
|
|
def _get_schedule_suspend_run_now(self, crawlconfig):
|
|
""" get schedule/suspend/run_now data based on crawlconfig """
|
|
|
|
# Create Cron Job
|
|
suspend = False
|
|
schedule = crawlconfig.schedule
|
|
|
|
if not schedule:
|
|
schedule = DEFAULT_NO_SCHEDULE
|
|
suspend = True
|
|
|
|
return suspend, schedule
|
|
|
|
async def _send_sig_to_pods(self, pods, aid):
|
|
command = ["kill", "-s", "SIGUSR1", "1"]
|
|
interrupted = False
|
|
|
|
try:
|
|
for pod in pods:
|
|
if pod.metadata.labels["btrix.archive"] != aid:
|
|
continue
|
|
|
|
await self.core_api_ws.connect_get_namespaced_pod_exec(
|
|
pod.metadata.name,
|
|
namespace=self.namespace,
|
|
command=command,
|
|
stdout=True,
|
|
)
|
|
interrupted = True
|
|
|
|
# pylint: disable=broad-except
|
|
except Exception as exc:
|
|
print(f"Exec Error: {exc}")
|
|
|
|
return interrupted
|
|
|
|
async def _delete_crawl_configs(self, label):
|
|
"""Delete Crawl Cron Job and all dependent resources, including configmap and secrets"""
|
|
|
|
await self.batch_beta_api.delete_collection_namespaced_cron_job(
|
|
namespace=self.namespace,
|
|
label_selector=label,
|
|
propagation_policy="Foreground",
|
|
)
|
|
|
|
await self.core_api.delete_collection_namespaced_config_map(
|
|
namespace=self.namespace,
|
|
label_selector=label,
|
|
propagation_policy="Foreground",
|
|
)
|
|
|
|
async def _create_run_now_job(self, cron_job):
|
|
"""Create new job from cron job to run instantly"""
|
|
annotations = cron_job.spec.job_template.metadata.annotations
|
|
annotations["btrix.run.manual"] = "1"
|
|
annotations["btrix.run.schedule"] = ""
|
|
|
|
# owner_ref = client.V1OwnerReference(
|
|
# kind="CronJob",
|
|
# name=cron_job.metadata.name,
|
|
# block_owner_deletion=True,
|
|
# controller=True,
|
|
# uid=cron_job.metadata.uid,
|
|
# api_version="batch/v1beta1",
|
|
# )
|
|
|
|
ts_now = datetime.datetime.utcnow().strftime("%Y%m%d%H%M%S")
|
|
name = f"crawl-now-{ts_now}-{cron_job.metadata.labels['btrix.crawlconfig']}"
|
|
|
|
object_meta = client.V1ObjectMeta(
|
|
name=name,
|
|
annotations=annotations,
|
|
labels=cron_job.metadata.labels,
|
|
# owner_references=[owner_ref],
|
|
)
|
|
|
|
job = client.V1Job(
|
|
kind="Job",
|
|
api_version="batch/v1",
|
|
metadata=object_meta,
|
|
spec=cron_job.spec.job_template.spec,
|
|
)
|
|
|
|
return await self.batch_api.create_namespaced_job(
|
|
body=job, namespace=self.namespace
|
|
)
|
|
|
|
def _get_job_template(
|
|
self,
|
|
cid,
|
|
storage_name,
|
|
storage_path,
|
|
labels,
|
|
annotations,
|
|
crawl_timeout,
|
|
parallel,
|
|
):
|
|
"""Return crawl job template for crawl job, including labels, adding optiona crawl params"""
|
|
|
|
requests_memory = "256M"
|
|
limit_memory = "1G"
|
|
|
|
requests_cpu = "120m"
|
|
limit_cpu = "1000m"
|
|
|
|
resources = {
|
|
"limits": {
|
|
"cpu": limit_cpu,
|
|
"memory": limit_memory,
|
|
},
|
|
"requests": {
|
|
"cpu": requests_cpu,
|
|
"memory": requests_memory,
|
|
},
|
|
}
|
|
|
|
job_template = {
|
|
"metadata": {"annotations": annotations},
|
|
"spec": {
|
|
"backoffLimit": self.crawl_retries,
|
|
"parallelism": parallel,
|
|
"template": {
|
|
"metadata": {"labels": labels},
|
|
"spec": {
|
|
"containers": [
|
|
{
|
|
"name": "crawler",
|
|
"image": self.crawler_image,
|
|
"imagePullPolicy": self.crawler_image_pull_policy,
|
|
"command": [
|
|
"crawl",
|
|
"--config",
|
|
"/tmp/crawl-config.json",
|
|
],
|
|
"volumeMounts": [
|
|
{
|
|
"name": "crawl-config",
|
|
"mountPath": "/tmp/crawl-config.json",
|
|
"subPath": "crawl-config.json",
|
|
"readOnly": True,
|
|
}
|
|
],
|
|
"envFrom": [
|
|
{"configMapRef": {"name": "shared-crawler-config"}},
|
|
{"secretRef": {"name": f"storage-{storage_name}"}},
|
|
],
|
|
"env": [
|
|
{
|
|
"name": "CRAWL_ID",
|
|
"valueFrom": {
|
|
"fieldRef": {
|
|
"fieldPath": "metadata.labels['job-name']"
|
|
}
|
|
},
|
|
},
|
|
{"name": "STORE_PATH", "value": storage_path},
|
|
{
|
|
"name": "STORE_FILENAME",
|
|
"value": "@ts-@hostname.wacz",
|
|
},
|
|
],
|
|
"resources": resources,
|
|
}
|
|
],
|
|
"volumes": [
|
|
{
|
|
"name": "crawl-config",
|
|
"configMap": {
|
|
"name": f"crawl-config-{cid}",
|
|
"items": [
|
|
{
|
|
"key": "crawl-config.json",
|
|
"path": "crawl-config.json",
|
|
}
|
|
],
|
|
},
|
|
}
|
|
],
|
|
"restartPolicy": "OnFailure",
|
|
},
|
|
},
|
|
},
|
|
}
|
|
|
|
if crawl_timeout > 0:
|
|
job_template["spec"]["activeDeadlineSeconds"] = crawl_timeout
|
|
|
|
return job_template
|