Very long (10 - 30 seconds) transaction failure timeout while offline

[larssn]

[READ] Step 1: Are you in the right place?

I was sent here by firebase/flutterfire#6736 (comment).

I've reproduced this in a Flutter project, but the guys over there thought it to be an underlying problem with the android SDK.

[REQUIRED] Step 2: Describe your environment

• Android Studio version: Using VSCode
• Firebase Component: Firestore
• Component version: 23.0.3

[REQUIRED] Step 3: Describe the problem

Steps to reproduce:

1. Setup a new project that has a button for a transaction that gets a document (doesnt matter which - we're offline).
2. Kill the internet for the tablet. It shouldn't matter how you do this.
3. Press the button and wait for the transaction to fail, and note the time it takes. Do this several times. It should fail quick at first (30ms - 50ms), but after 6ish clicks the time will jump to 10+ seconds. The times will get progressively worse until the tablet has "rested" for a few minutes, after which it returns to failing in milliseconds and the cycle starts over.

I hope those directions are sufficient without me having to set up a new environment. 😅

[aguatno]

Hi @ larssn thanks for reporting this issue. Though, I wasn't able to reproduce the issue using quickstart, could it be possible to provide a runnable mcve that we can use to replicate the issue? Thanks.

[larssn]

So you're saying your transactions always failed immediately while offline?

[aguatno]

HI @larssn Sorry, I was able to reproduce now. Allow me to consult this with our engineering team and will get back to you with updates.

[wu-hui]

Hi @larssn

Firestore transaction requires network to work, and has a re-try logic before it reports failure, hence the long failure time. We are also unable to detect network condition reliably, so fail early is also difficult to do. To the SDK, it might just be the network is not reliable.

I'd suggest you disable all operations that would lead to transactions, if your app can be sure it is offline.

[larssn]

@wu-hui @aguatno
I don't understand why this was closed, the issue remains.

I know it requires network to work, thats not the point of this issue. In the real world, networks can sporadically be unavailable.

Why does it fail immediately sometimes, while at others it takes 20+ seconds? That is inconsistent behaviour, which makes using the SDK needlessly complicated.

[wu-hui]

I see, the issue is that it is not consistent..not that it takes a while to fail.

I think this is because multiple transactions are initiated, and all of them are doing retries, on a single thread. This might lead to some of them being blocked. I will verify this.