Also affects juju 2.8.7-focal-amd64. Couple notes and a simplified use case; it seems the fan gets in the way of deployments that don't even leverage network-get:
$ juju add-model testu
Added 'testu' model on google/us-west1 with credential 'foo' for user 'admin'
Debug-log on a permanently-allocating unit looks like:
2021-02-04 23:10:30 INFO juju.agent.tools symlinks.go:20 ensure jujuc symlinks in /var/lib/juju/tools/unit-ubuntu-1
2021-02-04 23:10:30 INFO juju.agent.tools symlinks.go:40 was a symlink, now looking at /var/lib/juju/tools/2.8.7-bionic-amd64
2021-02-04 23:10:30 INFO juju.worker.uniter uniter.go:302 unit "ubuntu/1" started
2021-02-04 23:10:30 INFO juju.worker.uniter uniter.go:581 resuming charm install
2021-02-04 23:10:30 INFO juju.worker.uniter.charm bundles.go:79 downloading cs:ubuntu-17 from API server
2021-02-04 23:10:30 INFO juju.downloader download.go:111 downloading from cs:ubuntu-17
2021-02-04 23:26:23 INFO juju.worker.uniter uniter.go:286 unit "ubuntu/1" shutting down: preparing operation "install cs:ubuntu-17" for ubuntu/1: failed to download charm "cs:ubuntu-17" from API server: read tcp 252.0.135.1:44996->252.0.54.1:17070: read: connection reset by peer
2021-02-04 23:26:23 ERROR juju.worker.dependency engine.go:671 "uniter" manifold worker returned unexpected error: preparing operation "install cs:ubuntu-17" for ubuntu/1: failed to download charm "cs:ubuntu-17" from API server: read tcp 252.0.135.1:44996->252.0.54.1:17070: read: connection reset by peer
When re-deployed on a model with the workaround from comment #6 (fan-config="" container-networking-method=""), things are fine.
Also affects juju 2.8.7-focal-amd64. Couple notes and a simplified use case; it seems the fan gets in the way of deployments that don't even leverage network-get:
$ juju add-model testu
Added 'testu' model on google/us-west1 with credential 'foo' for user 'admin'
$ juju deploy ubuntu -n 5
Located charm "cs:ubuntu-17".
Deploying charm "cs:ubuntu-17".
$ sleep 1200 # shouldn't take 20m to get 5 ubuntus deployed, but let's be generous
$ juju status
Model Controller Cloud/Region Version SLA Timestamp
testu google-us-west1 google/us-west1 2.8.7 unsupported 17:26:57-06:00
App Version Status Scale Charm Store Rev OS Notes
ubuntu 18.04 waiting 2/5 ubuntu jujucharms 17 ubuntu
Unit Workload Agent Machine Public address Ports Message
ubuntu/0* active idle 0 35.230.121.28 ready
ubuntu/1 waiting allocating 1 35.233.238.175 agent initializing
ubuntu/2 waiting allocating 2 35.197.52.126 agent initializing
ubuntu/3 waiting allocating 3 35.233.131.212 agent initializing
ubuntu/4 active idle 4 34.82.186.17 ready
Debug-log on a permanently- allocating unit looks like:
2021-02-04 23:10:30 INFO juju.agent.tools symlinks.go:20 ensure jujuc symlinks in /var/lib/ juju/tools/ unit-ubuntu- 1 juju/tools/ 2.8.7-bionic- amd64 uniter. charm bundles.go:79 downloading cs:ubuntu-17 from API server 1:44996- >252.0. 54.1:17070: read: connection reset by peer dependency engine.go:671 "uniter" manifold worker returned unexpected error: preparing operation "install cs:ubuntu-17" for ubuntu/1: failed to download charm "cs:ubuntu-17" from API server: read tcp 252.0.135. 1:44996- >252.0. 54.1:17070: read: connection reset by peer
2021-02-04 23:10:30 INFO juju.agent.tools symlinks.go:40 was a symlink, now looking at /var/lib/
2021-02-04 23:10:30 INFO juju.worker.uniter uniter.go:302 unit "ubuntu/1" started
2021-02-04 23:10:30 INFO juju.worker.uniter uniter.go:581 resuming charm install
2021-02-04 23:10:30 INFO juju.worker.
2021-02-04 23:10:30 INFO juju.downloader download.go:111 downloading from cs:ubuntu-17
2021-02-04 23:26:23 INFO juju.worker.uniter uniter.go:286 unit "ubuntu/1" shutting down: preparing operation "install cs:ubuntu-17" for ubuntu/1: failed to download charm "cs:ubuntu-17" from API server: read tcp 252.0.135.
2021-02-04 23:26:23 ERROR juju.worker.
When re-deployed on a model with the workaround from comment #6 (fan-config="" container- networking- method= ""), things are fine.