Change script to support tensorflow distribution on k8s. #617
Conversation
docker/k8s_tools.py
Outdated
|
|
||
| r = [] | ||
| for ip in ips: | ||
| if port != "0": |
There was a problem hiding this comment.
Use port != None as it's the default value of the function arg.
docker/k8s_tools.py
Outdated
| pod_list = fetch_pods_info(label_selector) | ||
| pserver_ips = [item[1] for item in pod_list] | ||
| return ",".join(pserver_ips) | ||
| def fetch_ips(label_selector, port=None, phase=None): |
There was a problem hiding this comment.
This function is called fetch_ips but trying to append ports, try using fetch_endpoints
docker/k8s_tools.py
Outdated
| trainer_ips.sort() | ||
|
|
||
| def fetch_id(label_selector, phase=None): | ||
| pod_list = fetch_pods_info(label_selector, phase) |
There was a problem hiding this comment.
These lines are the same as fetch_ips, try put them in one function.
docker/k8s_tools.py
Outdated
| return fetch_endpoints(label_selector, port, phase) | ||
|
|
||
|
|
||
| def fetch_pserver_ips(label_selector, port=None, phase=None): |
There was a problem hiding this comment.
I'm confused fetch_pserver_ips called fetch_endpoints function, fetch_endpoints would return IP:PORT rather than IP.
There was a problem hiding this comment.
fetch_xxx_ips can have no argument port because we always use None. And, maybe it's better to add some comments to tell if port=None, fetch_endpoints will return a list of IPs.
docker/k8s_tools.py
Outdated
| api_response = v1.list_namespaced_pod( | ||
| namespace=NAMESPACE, pretty=True, label_selector=label_selector) | ||
| pod_list = [] | ||
| for item in api_response.items: | ||
| if phase is not None and item.status.phase != phase: |
There was a problem hiding this comment.
If phase == "Running" then need to check if it's Terminating
docker/k8s_tools.py
Outdated
| return fetch_endpoints(label_selector, port, phase) | ||
|
|
||
|
|
||
| def fetch_pserver_ips(label_selector, port=None, phase=None): |
There was a problem hiding this comment.
fetch_xxx_ips can have no argument port because we always use None. And, maybe it's better to add some comments to tell if port=None, fetch_endpoints will return a list of IPs.
No description provided.