Resolving merge conflicts.

Resolving merge conflict.

Adding of request cleaning.
Adding of session re-fetching after token update.
Resolving merge conflict.

Resolving merge conflict.

Error message changed, rebased on the latest changes.

Merging on recent changes. Changes to inspect first error item instead of iterating.

Merging on the latest updates.

Merged with the recent changes.

Rebasing.

Fix merging with the latest changes.

Rebasing.

Merging nested if's.

Removing return of error if the device token is empty.

In D5210#153347, @tomek wrote:

Shouldn't we check if device is online?

Adding of checking for device is online.
Error message text was changed.

In D5174#153312, @tomek wrote:

I'm not really sure about the architecture here. Why is ping a part of get method? Shouldn't we start sending pings when a new session is created? Maybe I don't see something or don't understand the requirements, but doing pings in get sounds a bit useless - we're sending other messages there and based on the result can determine if a client is online. Pinging mechanism should work even when there is no other communication.

Maybe I'm missing something. @max could you explain why we're doing this in get? Is there something wrong in starting pinging when a new session is established?

Removing of the if condition.

In D5205#152866, @varun wrote:

ah thanks. do we actually use these files anywhere though?

In D5145#153069, @ashoat wrote:

@max, can you rebase on master to see if the CI still fails?

In D5173#153063, @ashoat wrote:

Probably @max didn't rebase on master after @jon's recent changes.

Rebasing on a latest master.

Making topic not optional.

In D5083#149217, @tomek wrote:

This diff is confusing. You're mentioning that

To wake up the iOS app from the background we should send a message with the payload. In this case, a message will be with the content-available attribute, so we can do background information processing in the app.

But this diff doesn't set this flag anywhere.

Changes to change a content-available flag itself depsite adding of the payload_data.

Adding a comment to explain why we are returning Ok with error types.

Adding a comment to explain why we are returning Ok with error types.

Changes to return bool result instead of void.

In D5173#152652, @ashoat wrote:

We'll rethink this API in ENG-1072 and ENG-484. Would be good to prioritize ENG-1842 for Tunnelbroker v0.5 as well... I would like us to nail the API down in that version, so we can avoid having to change it while the service is live in the future.

In D5175#151890, @jon wrote:

would be nice to have a test case that the online behavior is being correctly updated.

In D5175#152307, @tomek wrote:

This is rather strange: for each get request we have an infinite loop of ping - is this the right approach?

Rebasing on the parent changes.

Rebased on using the empty instead of bool for the ping.
Using of TryCancel() was removed in a favor of the atomic_bool.
Some small code reordering.

In D5173#152584, @tomek wrote:

Not really sure about good practices here, but for me it sounds strange that we're using the same function for two completely different purposes. Shouldn't we introduce a new function which handles just the ping?

Change ping type from bool to google.protobuf.Empty.

messageResponse changed to responseMessage to follow a convention. Fixing a typo.

Fixing native client build.

In D5164#152455, @ashoat wrote:

Tunnelbroker unit tests seem to be failing?

Adding @ashoat as a blocking reviewer, because this diff touches dependencies.