Be aware that both approaches described above (serialize/deserialize by tuk and custom converter by Zarnuk) will produce different outputs.
With the serialize/deserialize approach:
- Field names in snake_case format will be automatically converted into camelCase.
JsonFormat.printer() does this.
- Numeric values will be converted to float.
Gson does that for you.
- Values of type Duration will be converted into strings with format durationInseconds + "s", i.e. "30s" for a duration of 30 seconds and "0.000500s" for a duration of 500,000 nanoseconds.
JsonFormat.printer() does this.
With the custom converter approach:
- Field names will remain as they are described on the proto file.
- Integers and floats will keep their own type.
- Values of type Duration will become objects with their corresponding fields.
To show the differences, here is a comparison of the outcomes of both approaches.
Original message (here is the proto file):
method_config {
name {
service: "helloworld.Greeter"
method: "SayHello"
}
retry_policy {
max_attempts: 5
initial_backoff {
nanos: 500000
}
max_backoff {
seconds: 30
}
backoff_multiplier: 2.0
retryable_status_codes: UNAVAILABLE
}
}
With the serialize/deserialize approach:
{
methodConfig=[ // field name was converted to cameCase
{
name=[
{
service=helloworld.Greeter,
method=SayHello
}
],
retryPolicy={
maxAttempts=5.0, // was integer originally
initialBackoff=0.000500s, // was Duration originally
maxBackoff=30s, // was Duration originally
backoffMultiplier=2.0,
retryableStatusCodes=[
UNAVAILABLE
]
}
}
]
}
With the custom converter approach:
{
method_config=[ // field names keep their snake_case format
{
name=[
{
service=helloworld.Greeter,
method=SayHello
}
],
retry_policy={
max_attempts=5, // Integers stay the same
initial_backoff={ // Duration values remains an object
nanos=500000
},
max_backoff={
seconds=30
},
backoff_multiplier=2.0,
retryable_status_codes=[
UNAVAILABLE
]
}
}
]
}
Bottom line
So which approach is better?
Well, it depends on what you are trying to do with the Map<String, ?>. In my case, I was configuring a grpc client to be retriable, which is done via ManagedChannelBuilder.defaultServiceConfig API. The API accepts a Map<String, ?> with this format.
After several trials and errors, I figured that the defaultServiceConfig API assumes you are using GSON, hence the serialize/deserialize approach worked for me.
One more advantage of the serialize/deserialize approach is that the Map<String, ?> can be easily converted back to the original protobuf value by serializing it back to json, then using the JsonFormat.parser() to obtain the protobuf object:
ServiceConfig original;
...
String asJson = JsonFormat.printer().print(original);
Map<String, ?> asMap = new Gson().fromJson(asJson, Map.class);
// Convert back to ServiceConfig
String backToJson = new Gson().toJson(asMap);
ServiceConfig.Builder builder = ServiceConfig.newBuilder();
JsonFormat.parser().merge(backToJson, builder);
ServiceConfig backToOriginal = builder.build();
... whereas the custom converter approach method doesn't have an easy way to convert back as you need to write a function to convert the map back to the original proto by navigating the tree.
