/layer-etcd

WIP - Working multi-snap cluster #77

Closed
wants to merge 7 commits into
from

Conversation

Reviewers

@battlemidget battlemidget

@marcoceppi marcoceppi

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
3 participants
@chuckbutler @battlemidget @marcoceppi
@chuckbutler
Collaborator

chuckbutler commented Feb 19, 2017 •
Edited 1 time

DO NOT MERGE!

This is an open thread to track the progress and discuss the approach to the refactoring of the etcd layer to migrate from apt delivery => snaps.

This has implications on:

  • delivery mechanism
  • backup/restore
  • durable storage migration(s)

Initial deployment + upgrade code. Functional X node cluster. Does not survive leader termination.
@chuckbutler
WIP - Working multi-snap cluster 
Initial deployment (non-upgrade) code. Functional X node cluster. Does not survive leader termination.
e7c8890
templates/defaults
@@ -1,5 +1,5 @@
# This file is rendered by Juju, manual edits will not be persisted
-ETCD_DATA_DIR=/var/lib/etcd/default
+ETCD_DATA_DIR={{ etcd_data_dir }}/{{ unit_name }}.etcd
@chuckbutler

chuckbutler Feb 19, 2017

Collaborator

This now matches what the snap is doing. I'm not 100% positive we want to keep this nomenclature but I'm for keeping it in lockstep with the snap so it stays consistent.

reactive/etcd.py
+ else:
+ subject = "peer"
+
+ status_message = "{0} with {1} known {2}.".format(unit_health, peers,
@marcoceppi

marcoceppi Feb 20, 2017

Owner

Could we do this instead?

status_message = '{0} with {1} known peer{2}'.format(unit_health, peers, 's' if peers != 1 else '')
@chuckbutler

chuckbutler Feb 20, 2017

Collaborator

We could, I don't know what it buys us other than collapsing four lines into one, but at the sake of readability. I don't feel strongly either way however.

actions/snap-upgrade
+ else
+ # Install from the resource location, and check for correct snap.
+ MANUAL_SNAP=$(resource-get etcd)
+ snap install ${ETCD_MANUAL_SNAP} --classic
@battlemidget

battlemidget Feb 21, 2017

If you are installing a local snap you need to pass --dangerous as well

actions/snap-upgrade
+ snap refresh etcd --channel=${CONFIGURED_CHANNEL} --classic
+else
+ MANUAL_SNAP=$(resource-get etcd)
+ snap refresh ${MANUAL_SNAP}
@battlemidget

battlemidget Feb 21, 2017

Also i dont think snap refresh works for locally installed snaps

@marcoceppi

marcoceppi Feb 24, 2017

Owner

You'll need to just snap install /again/

@chuckbutler chuckbutler referenced this pull request in juju-solutions/bundle-canonical-kubernetes Feb 22, 2017

Closed

Migration to k8s v1.6 and etcd v3 #208

chuckbutler added some commits Feb 24, 2017

@chuckbutler
Actions snap refactor
131efcc
@chuckbutler
Tweaked peer health string
468ba19
@chuckbutler
Adds etcd3 snap support (prelim) 
Tested with 3.1.2.

- Refactors template rendering to the version listed as installed in
snapd.
- Adds etcd2/3 configuration templates
- Readme updates
- Removing amulet package from test dependencies.
26a43e6
@chuckbutler
Cleanup of restore and snap-upgrade work 
- Move "migrate.py" work to snap-upgrade
- Refactores restore to be python based and works with both 2.x and 3.x
series.
- Changes default casing in actions.yaml bool params.
27b32a4
@chuckbutler
Refactor storage to work with $SNAP_DATA 
Storage was historically temp mounted, migrated, then blindly mounted
over the default $ETCD_DATA path. This now refactores the storage to
work with snaps that have access to the 'pluggable-media' interface.

This particular feature depends on
markshuttle/etcd-snaps#4
66461a3
@chuckbutler
Collaborator

chuckbutler commented Mar 9, 2017

Closing for a formal PR

@chuckbutler chuckbutler closed this Mar 9, 2017

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment