Exosphere

@kageurufu observed there is also an issue with setting the instance passphrase. Once this is done we should not attempt to do it again, even across subsequent Ansible runs. So, we need a way to track this.

There is a new version: https://github.com/TurboVNC/turbovnc/releases/tag/3.1.3

@kageurufu concern: re-running a partially finished Ansible playbook could result in bad state on the instance (due to inadequate idempotence). So for now, we should probably only retry the git repository download, and run the Ansible only once. ... [More] We should create a separate issue to track work to make the Ansible code more idempotent. Frank has tried and found unexpected dependencies between roles. [Less]

Likely mitigated with newer version of Guacamole; Frank asked about timing of TurboVNC release with upstream fix.

retry git clone \\ --depth=1 \\ --branch="{instance-config-mgt-repo-checkout}" \\ "{instance-config-mgt-repo-url}" \\ /opt/instance-config-mgt ansible-playbook \\ -i /opt/instance-config-mgt/ansible/hosts \\ -e "{ansible-extra-vars}" \\ /opt/instance-config-mgt/ansible/playbook.yml

This works, thanks. MR now needs approval.

Separating this out of #715. Problem/Opportunity Statement The create instance page is essentially a web form with several fields. Sometimes, a workshop that uses Exosphere will require each student or participant to create their own instance ... [More] , configured on the create instance page in a specific way. This can be tedious and error-prone, especially if someone needs to edit (e.g.) the cloud-init user data. What would success / a fix look like? Exosphere can allow workshop facilitators to generate a special URL and send it to participants. This URL pre-populates the interface with correct values which exposes the instance (or data science workbench) setup process to the participant, while reducing potential for user-entered mistakes. Perhaps we achieve this with a long query string. It may need to include cloud-init metadata. (What is max length of query string?) This needs some design work. [Less]

Overview Implements retry logic for the parts of instance setup that require network connectivity to external services, like downloading and running the Ansible playbook. If something fails, we wait a moment and try again, up to 3 times, before ... [More] failing outright. Fixes #1021 (or at least works around it from Exosphere's perspective). This also switches our cloud-init data format to a MIME multi-part archive. Mostly, I did this to rescue the bash code out of a YAML blob, making said code easier to develop and troubleshoot. Additionally, this is forward-compatible with solutions to #919 and #701, where we allow the user or deployer (respectively) to specify their own boot scripts (or cloud-init user data), which we then combine with Exosphere's own cloud-init user data as a MIME multi-part archive. How to Test Perhaps, briefly break an instance's network connectivity while it's running setup (perhaps by removing the egress rule in its security group). Watch the instance console log (or /var/log/cloud-init-output.log to see the script retry. Screenshots No UI changes. [Less]

Overview Implements retry logic for the parts of instance setup that require network connectivity to external services, like downloading and running the Ansible playbook. If something fails, we wait a moment and try again, up to 3 times, before ... [More] failing outright. Fixes #1021 (or at least works around it from Exosphere's perspective). This also switches our cloud-init data format to a MIME multi-part archive. Mostly, I did this to rescue the bash code out of a YAML blob, making said code easier to develop and troubleshoot. Additionally, this is forward-compatible with solutions to #919 and #701, where we allow the user or deployer (respectively) to specify their own boot scripts (or cloud-init user data), which we then combine with Exosphere's own cloud-init user data as a MIME multi-part archive. How to Test Perhaps, briefly break an instance's network connectivity while it's running setup (perhaps by removing the egress rule in its security group). Watch the instance console log (or /var/log/cloud-init-output.log to see the script retry. Screenshots No UI changes. [Less]

Chris Martin (81823057) at 06 Nov 23:56 Merge branch 'cloud-init-multipart-retry' into 'master' ... and 3 more commits