Durable database writes

[wcalderw]

We need a way to force data writes to disk to prevent loss of data in the event the app/database is "stopped". Firefox has implemented an experimental mode "readwriteflush" that can be used on a transaction such that when oncomplete event happens with success you can be sure that the data made it to the disk. This circumvents the "write cache" that has been implemented in Indexeddb and impacts performance, but for those of use that must ensure data is permanently stored this is a trade off we are willing to have.

As it stands. We have observed Edge Windows 10 1607 (OS level write caching turned off) IndexedDB can take 1 to 15+ minutes before committing a write to disk. We are testing with other systems (Android, IOS, etc) as well, but suspect the same behavior.

[beidson]

WebKit is certainly not interested in adding any new web APIs that involve synchronous i/o, which is basically what is being asked for here.

Are you actually finding in other implementations that after a transaction commits, there's a 15 minute window where the bits are not written to disk?

That sounds like an issue to raise with those implementations specifically.

[inexorabletash]

For reference, Chrome does not fire complete until the transaction has flushed to disk. This happens in a different process from the page so does not block any event loops.

[wcalderw]

We are still investigating other platforms (Android Chromium and IOS WebKit). If chrome is actually generating the complete as suggested then at least that may solve one of the 3 platforms for us. Windows has been a real pain as far as app limitations and short commings... I'll follow up here after we've looked at android.

[inexorabletash]

@wcalderw - any feedback?

[wcalderw]

We moved to a cordova provided no sql solution on windows 10 to avoid the problem. Was really frustrating to have a success on a data submission only to find it's not really saved though... Seems like a non starter for a database solution.

…

On Fri, Jan 13, 2017 at 7:30 PM, Joshua Bell ***@***.***> wrote: @wcalderw <https://github.com/wcalderw> - any feedback? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#99 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AB8h_WjT1kOACmdwhsCnQbgV0B3pSNsPks5rSBcsgaJpZM4KcI3h> .

[wcalderw]

Oh and we still haven't received any response from Microsoft on this. Despite them making our submitted issue critical. On Fri, Jan 13, 2017 at 9:31 PM, William Calderwood <[email protected]> wrote:

…

We moved to a cordova provided no sql solution on windows 10 to avoid the problem. Was really frustrating to have a success on a data submission only to find it's not really saved though... Seems like a non starter for a database solution. On Fri, Jan 13, 2017 at 7:30 PM, Joshua Bell ***@***.***> wrote: > @wcalderw <https://github.com/wcalderw> - any feedback? > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub > <#99 (comment)>, or mute > the thread > <https://github.com/notifications/unsubscribe-auth/AB8h_WjT1kOACmdwhsCnQbgV0B3pSNsPks5rSBcsgaJpZM4KcI3h> > . >

[inexorabletash]

For performance reasons, we (Chrome) are considering making the flush optional. Example:

db.transaction(scope, "readwrite", {flush: false});

(Having the default be truthy is annoying, but there's no good antonym. Suggestions?)

Thoughts?

[domenic]

preventFlush: true?

[inexorabletash]

Maybe. Feels like a double-negative. ref

[domenic]

We've done it a few times, e.g. .focus({ preventScroll: true }).

[inexorabletash]

I guess this is a dupe of #50 ... let me close it out and move the comments there.