Syncing Upstream Updates

Updating the facade and definitions code generated from the schema to reflect changes in upstream Juju consists of two steps:

  • Creating a new schemas-juju-<version>.json file from the Juju code-base
  • Generating the libjuju Python code from that schema

Additionally changes, although rare may be also required to add or update an override.

Creating a Schema File

Firstly, checkout the Juju project and correctly set it up according to it’s documentation. Juju has the ability to generate the schema and stores it in version control (git). Copy that file from Juju to libjuju repository to make sure it can be picked up for construction of the libjuju client.

From inside Juju:

make rebuild-schema

The version number used in the filename should match the upstream version of Juju. To ensure that the client picks up the new change, the latest pointer (schemas-juju-latest.json) should be pointed to the versioned file:

From inside libjuju:

cp ${GOPATH}/src/github.com/juju/juju/apiserver/facades/schema.json \
    juju/client/schemas-juju-2.6.4.json
rm juju/client/schemas-juju-latest.json
ln -s schemas-juju-2.6.4.json juju/client/schemas-juju-latest.json

Generating the Python Code

Creating a new client requires the building of the Python code from the schema and doing so is a make target:

make client

Changes to juju/client/_definitions.py and juju/client/_clientX.py files are expected, along with any facades files, depending what was changed in Juju.

Integrating into the Object Layer

Once the raw client APIs are synced, you may need to integrate any new or changed API calls into the object layer, to provide a clean, Pythonic way to interact with the model. This may be as simple as adding an optional parameter to an existing model method, tweaking what manipulations, if any the model method does to the data before it is sent to the API, or it may require adding an entirely new model method to capture the new functionality.

In general, the approach should be to make the interactions with the model layer use the same patterns as when you use the CLI, just with Python idioms and OO approaches.

When trying to determine what client calls need to be made and what data to be sent for a given Juju CLI action, it is very useful to add –debug –logging-config TRACE to any Juju CLI command to view the full conversation between the CLI client and the API server. For example:

` [johnsca@murdoch:~] $ juju deploy --debug --logging-config TRACE ./builds/test 11:51:20 INFO  juju.cmd supercommand.go:56 running juju [2.3.5 gc go1.10] 11:51:20 DEBUG juju.cmd supercommand.go:57   args: []string{"/snap/juju/3884/bin/juju", "deploy", "--debug", "--logging-config", "TRACE", "./builds/test"} 11:51:20 INFO  juju.juju api.go:67 connecting to API addresses: [35.172.119.191:17070 172.31.94.16:17070 252.94.16.1:17070] 11:51:20 TRACE juju.api certpool.go:49 cert dir "/etc/juju/certs.d" does not exist 11:51:20 DEBUG juju.api apiclient.go:843 successfully dialed "wss://35.172.119.191:17070/model/a7317969-6dab-4ba4-844b-af3d661c228d/api" 11:51:20 INFO  juju.api apiclient.go:597 connection established to "wss://35.172.119.191:17070/model/a7317969-6dab-4ba4-844b-af3d661c228d/api" ... 11:51:20 INFO  juju.cmd.juju.application series_selector.go:71 with the configured model default series "xenial" 11:51:20 DEBUG httpbakery client.go:244 client do POST https://35.172.119.191:17070/model/a7317969-6dab-4ba4-844b-af3d661c228d/charms?revision=0&schema=local&series=xenial { 11:51:21 DEBUG httpbakery client.go:246 } -> error <nil> 11:51:21 INFO  cmd deploy.go:1096 Deploying charm "local:xenial/test-0". 11:51:21 TRACE juju.rpc.jsoncodec codec.go:225 -> {"request-id":3,"type":"Charms","version":2,"request":"CharmInfo","params":{"url":"local:xenial/test-0"}} 11:51:21 TRACE juju.rpc.jsoncodec codec.go:120 <- {"request-id":3,"response":{"revision":0,"url":"local:xenial/test-0","config":{"test":{"type":"string","default":""}},"meta":{"name":"test","summary":"test","description":"test","subordinate":false,"series":["xenial"],"resources":{"dummy":{"name":"dummy","type":"file","path":"dummy.snap","description":"dummy snap"}},"min-juju-version":"0.0.0"},"actions":{}}} 11:51:21 TRACE juju.rpc.jsoncodec codec.go:225 -> {"request-id":4,"type":"Charms","version":2,"request":"IsMetered","params":{"url":"local:xenial/test-0"}} 11:51:21 TRACE juju.rpc.jsoncodec codec.go:120 <- {"request-id":4,"response":{"metered":false}} 11:51:21 TRACE juju.rpc.jsoncodec codec.go:225 -> {"request-id":5,"type":"Resources","version":1,"request":"AddPendingResources","params":{"tag":"application-test","url":"local:xenial/test-0","channel":"","macaroon":null,"resources":[{"name":"dummy","type":"file","path":"dummy.snap","description":"dummy snap","origin":"store","revision":-1,"fingerprint":"","size":0}]}} 11:51:21 TRACE juju.rpc.jsoncodec codec.go:120 <- {"request-id":5,"response":{"pending-ids":["c0ffdd92-da23-4fb2-8d41-d82d58423447"]}} 11:51:21 TRACE juju.rpc.jsoncodec codec.go:225 -> {"request-id":6,"type":"Application","version":5,"request":"Deploy","params":{"applications":[{"application":"test","series":"xenial","charm-url":"local:xenial/test-0","channel":"","num-units":1,"config-yaml":"","constraints":{},"resources":{"dummy":"c0ffdd92-da23-4fb2-8d41-d82d58423447"}}]}} 11:51:21 TRACE juju.rpc.jsoncodec codec.go:120 <- {"request-id":6,"response":{"results":[{}]}} 11:51:21 TRACE juju.rpc.jsoncodec codec.go:123 <- error: read tcp 192.168.1.102:52168->35.172.119.191:17070: use of closed network connection (closing true) 11:51:21 DEBUG juju.api monitor.go:35 RPC connection died 11:51:21 INFO  cmd supercommand.go:465 command finished `

Note that this will contain login information (which has been removed from the above).

Overrides

It should be quite rare, but occasionally the generated Python code does not capture all of the logic needed to properly parse the output from the API or may otherwise need some small amount of tweaking. This is what the juju/client/overrides.py file is for. An example of this is the Number type, which isn’t standard JSON and must be parsed slightly differently.

At the top of that file are two lists, __all__ and __patches__. The former replaces entire class implementations, while the latter patches the attributes of the override classes into the matching generated class, leaving the rest of the generated class untouched.