sanxiyn ( Seo Sanghyeon )

In #8, lease keepalive task was added. The intent was to preserve the lease between lock request and lock acquisition. By mistake, the lease keepalive continued after the lock acquisition, making the lock TTL ineffective. Therefore, cancel the lease keepalive task when the lock is acquired.

Also add a new test. In test_lock_ttl_2, lock with TTL expires at T+4 without keepalive, T+6 with this PR, and unlocks without expiration at T+8 before this PR.

+54 -8

1 commit in lablup/etcetra/tree/fix/lock-ttl lablup/etcetra/tree/main

0 comment

2 changed files

pr created time in 2 months

create barnchlablup/etcetra

branch : fix/lock-ttl

created branch time in 2 months

push eventlablup/backend.ai

Seo Sanghyeon

commit sha d5dd554caa20d2125beab49d3ebb119054dc0845

Execute namespace dependent operation in the new process

view details

push time in 2 months

Pull request review commentlablup/backend.ai

feat: Gather per-container network statistics

 async def sysfs_impl(container_id):                     e,                 )                 return None+            async with closing_async(Docker()) as docker:+                container = DockerContainer(docker, id=container_id)+                data = await container.show()+                sandbox_key = data["NetworkSettings"]["SandboxKey"]+            net_rx_bytes = 0+            net_tx_bytes = 0+            with nsenter(sandbox_key):+                nstat = psutil.net_io_counters(pernic=True)

Rather, this should be executed in the new process... See my new commit and comment there for details.

sanxiyn

comment created time in 2 months

PullRequestReviewEvent

Pull request review commentlablup/backend.ai

feat: Gather per-container network statistics

 async def sysfs_impl(container_id):                     e,                 )                 return None+            async with closing_async(Docker()) as docker:+                container = DockerContainer(docker, id=container_id)+                data = await container.show()+                sandbox_key = data["NetworkSettings"]["SandboxKey"]

I created a new issue for this: #1196.

sanxiyn

comment created time in 2 months

PullRequestReviewEvent

issue openedlablup/backend.ai

Index agent's kernel registry by backend-dependent key

Currently, agent's kernel registry is indexed by kernel_id, which is the backend-independent key. Backends are not aware of it, for example, Docker backend only knows about container_id of its containers.

Sometimes we need to look up kernel by backend-dependent key. Currently, we do this by creating a temporary index by iterating over entire kernel registry. kernel_id_map: Dict[ContainerId, KernelId] in stats.py is an example. Instead, we should make this index permanent and update it whenever kernel registry is updated.

created time in 2 months

PullRequestReviewEvent

Pull request review commentlablup/backend.ai

feat: Gather per-container network statistics

 async def sysfs_impl(container_id):                     e,                 )                 return None+            async with closing_async(Docker()) as docker:+                container = DockerContainer(docker, id=container_id)+                data = await container.show()+                sandbox_key = data["NetworkSettings"]["SandboxKey"]

Sure.

sanxiyn

comment created time in 3 months

Pull request review commentlablup/backend.ai

feat: Gather per-container network statistics

 async def sysfs_impl(container_id):                     e,                 )                 return None+            async with closing_async(Docker()) as docker:+                container = DockerContainer(docker, id=container_id)+                data = await container.show()+                sandbox_key = data["NetworkSettings"]["SandboxKey"]+            net_rx_bytes = 0+            net_tx_bytes = 0+            with nsenter(sandbox_key):+                nstat = psutil.net_io_counters(pernic=True)

Oops, you are right. This should be executed in the new thread.

sanxiyn

comment created time in 3 months