blablacar / dgr Goto Github PK

2 ways :

pod:
  apps:
    - image: aci/aci
      app:
       ....
      dependencies:
          - deps/deps1
          - deps/deps2
          - deps/deps3

or

pod:
  apps:
    - from: 
         - aci/aci1
         - aci/aci2
         - aci/aci3
      app:
       ....

check changes in

check all files in build directory
check manifest

Like --target-path but which works for pod too (ie. multi target paths).

Maybe you can keep this option which would create subfolders for pod?

since we're only using confd for templating and extend templating capabilities for our needs.

Add the possibility to add an args to the command.
cnt init my/path

Posibly add a .gitignore and a default cnt-manifest file

I created a attributes directory with some galera config on my environment:

vi ../env/prod/services/db-log/attributes/galera.cnf.yml

---
override:
  galera:
    mysqld:
      wsrep_provider_options:
        gcache.size: "1G"
        gmcast.segment: 1
        gcs.fc_limit: 500
        gcs.fc_master_slave: "YES"
        gcs.fc_factor: "1.0"

It is basically some "environment" attributes.

I added some "node attributes" in the manifest as:

vi ../env/prod/services/db-log/service-manifest.yml

...
nodes:
  - hostname: mydb5
    ip: XX.XX.XX.XX
    fleet:
      - MachineMetadata="rack=1" "pos=1"
    attributes:
      galera:
        mysqld:
          wsrep_cluster_address:"gcomm://mydb1.prod,mydb2.prod,mydb3.prod,mydb4prod"

My problem is that the mysqld section is overrided by the node attribute.

In the unit I have:

[...]"galera":{"mysqld":"wsrep_cluster_address:\\\"gcomm://mydb1.prod,mydb2.prod,mydb3.prod,mydb4prod\\\""}[...]

Instead of

[...]"galera":{"mysqld":"wsrep_cluster_address:"gcomm://mydb1.prod,mydb2.prod,mydb3.prod,mydb4prod","wsrep_provider_options":{"gcache.size":"1G","gcs.fc_factor":"1.0","gcs.fc_limit":500,"gcs.fc_master_slave":"YES","gmcast.segment":1}}[...]

Note that the attribute wsrep_cluster_address is part of the mysqld value,

Key => "mysqld": 
Value => "wsrep_cluster_address:\\\"gcomm://bbcdblog4.prod.par-1.h.blbl.cr,bbcdblog5.prod.par-1.h.blbl.cr,bbcdblog6.prod.par-1.h.blbl.cr,dblog8.prod.par-3.h.blbl.cr\\\""

Thanks

We should be able to build without the flag --share-system for systemd-nspawn.
That would avoid the build script to mess with the build host.

If I'm not mistaken, cnt install should build as well, and produce a target/pod-manifest.json

currently we have to run :
cnt -c build
cnt install

cnt should apply gzip on acis only on push so testing would be faster

we should reimplement a pure go version that will handle the fact that the image already exists

remove eventHandler and support all new cnt standards.
and support pods

when running sudo with a tty, you want to run directly cnt as root and thus when to use the root home folder as cnt_home. Currently, since the SUDO_USER still exists, cnt use this using home folder.

according to the appc spec, an aci should only contain a rootfs and a manifest

as for now we can build a pod with a unexistent version of a dependency.
cnt should probably at least warn or even refuse to build.

hi
we should add a template for the manifest so it would be easier to start to write an aci.

Possibly a good idea to warn the users that they are pushing a version from which dependencies versions are not pushed yet.

example :

• build base:42
• (forget to push base:42)
• build myAci:2 with a dependency on base:42
• push myAci:2
• cnt WARNs : myAci:2 depends on base:42 which does not appear on image repository.

It could be a simple warn or even refuse to push. That way nobody can run an image with missing deps.

Validate manifest. For example projectName structure and not null