1346796 - Investigate why testing on bug 1341097 found more jank on 53b1 than on 54a2 when importing IuliaC's history file

Status: RESOLVED FIXED

(Firefox :: Migration, defect)

Description

[:Gijs (he/him)]

Context: bug 1341097 comment 49 and later.

We should figure out if the limit/sorting is having an effect here, or if there's some other difference (sqlite version?) that is causing the significant perf difference.

Comment 1

[:Gijs (he/him)]

I can't reproduce the difference locally, with an up-to-date (54.0a2 (2017-03-13) (64-bit)) devedition build and a separately downloaded copy of 53b1 (also 64-bit), on a clean profile. Here are my steps:

1. download build
2. run build with -p -no-remote
3. create a new profile and start with that new profile
4. click through the 'make this the default browser' dialog and/or startup tour stuff for devedition
5. close all the standard tabs that open, open a clean new tab
6. hit ctrl-shift-b to open the library
7. start the import wizard from the button on the library toolbar, select Chrome
8. untick passwords, bookmarks, cookies, then click import, and wait for the migrator to finish
9. check numbers in about:telemetry.

My numbers are also much lower than the ones from bug 1341097 comment #49, probably because my machine is much faster.

Iulia, are you doing something differently? Did I misunderstand something? Can you still reproduce the same effect?

Comment 2

[Iulia Cristescu, QA [:JuliaC] (away, please needinfo? [e-mail])]

(In reply to :Gijs from comment #1)
> I can't reproduce the difference locally, with an up-to-date (54.0a2
> (2017-03-13) (64-bit)) devedition build and a separately downloaded copy of
> 53b1 (also 64-bit), on a clean profile. Here are my steps:
> 
> 1. download build
> 2. run build with -p -no-remote
> 3. create a new profile and start with that new profile
> 4. click through the 'make this the default browser' dialog and/or startup
> tour stuff for devedition
> 5. close all the standard tabs that open, open a clean new tab
> 6. hit ctrl-shift-b to open the library
> 7. start the import wizard from the button on the library toolbar, select
> Chrome
> 8. untick passwords, bookmarks, cookies, then click import, and wait for the
> migrator to finish
> 9. check numbers in about:telemetry.
> 
> My numbers are also much lower than the ones from bug 1341097 comment #49,
> probably because my machine is much faster.
> 
> Iulia, are you doing something differently? Did I misunderstand something?
> Can you still reproduce the same effect?

Tried again on 54.0a2 (2017-03-13) (Windows 10 x64) , but this time closely following your steps. These are the results:
chrome
1 samples, average = 478, sum = 478

146 |  0  0%
267 |#########################  1  100%
487 |  0  0%

For bug 1341097 comment #49 I was running the same steps, except that I also imported the passwords, bookmarks and cookies in addition to the history. Maybe this made that big difference.

Comment 3

[Iulia Cristescu, QA [:JuliaC] (away, please needinfo? [e-mail])]

Also, using the same Chrome profile and the above steps, this is the result for 53.0b2 build1 (20170313154936):
chrome
1 samples, average = 439, sum = 439

146 |  0  0%
267 |#########################  1  100%
487 |  0  0%

Comment 4

[:Gijs (he/him)]

(In reply to Iulia Cristescu, QA [:IuliaC] from comment #3)
> Also, using the same Chrome profile and the above steps, this is the result
> for 53.0b2 build1 (20170313154936):
> chrome
> 1 samples, average = 439, sum = 439
> 
> 146 |  0  0%
> 267 |#########################  1  100%
> 487 |  0  0%

OK, so this is more or less the same, which is a lot better... I'm still puzzled about the difference in bug 1341097 though. If you follow the steps from comment #2 with both builds, without unticking cookies/bookmarks/passwords, what results do you get? Do you see similar numbers as in bug 1341097 in that case? If so, can you try to work out (by trying the different combinations of items) what's causing the difference? We know there are edgecases where we might try to run multiple migrations at the same time (see bug 1338800) but the last time I looked at the Chrome profile migrator I thought this wouldn't happen in practice. It's possible that I misread the code, though, in which case it would still be good to know what resource type is causing us to be so janky when importing (and why it's gotten worse in 54...).

Comment 5

[Iulia Cristescu, QA [:JuliaC] (away, please needinfo? [e-mail])]

(In reply to :Gijs from comment #4)
> If you follow the steps
> from comment #2 with both builds, without unticking
> cookies/bookmarks/passwords, what results do you get? 
Yes, I tested this scenario on 54.0a2 (2017-03-14) and 53.0b2 build1 (20170313154936) repeatedly, each time using a new profile. FX_MIGRATION_HISTORY_JANK_MS doesn't have quite constant values, even if running exactly the same steps every time. Here you have the values for each try https://docs.google.com/document/d/1Am9S0tMG1nptUx58dUeZe3jU8VbfXle-qmm4MHn_eAw/edit?usp=sharing. Also, if following the same steps but using the installed builds, the FX_MIGRATION_HISTORY_JANK_MS values seem to slightly increase. See https://docs.google.com/document/d/10LpkRqN_Ns5YVmLFmwr2f1kPi1B8gbfopelrwZtlFfI/edit?usp=sharing.  

> Do you see similar
> numbers as in bug 1341097 in that case? If so, can you try to work out (by
> trying the different combinations of items) what's causing the difference?
Using latest aurora and beta builds, I didn't manage to reproduce similar FX_MIGRATION_HISTORY_JANK_MS values with those from https://bugzilla.mozilla.org/show_bug.cgi?id=1341097#c49, but only with 53.0b1 build1 (20170307064827). Those were encountered for complete migration and for bookmarks & history migration cases. See https://docs.google.com/document/d/1R_00m69xUtGzaw6UC2vOKRzcfzfvpPl6ls9z8pIihyI/edit?usp=sharing. 

Please let me know if there is anything else to investigate.

Comment 6

[:Gijs (he/him)]

(In reply to Iulia Cristescu, QA [:IuliaC] from comment #5)
> (In reply to :Gijs from comment #4)
> > If you follow the steps
> > from comment #2 with both builds, without unticking
> > cookies/bookmarks/passwords, what results do you get? 
> Yes, I tested this scenario on 54.0a2 (2017-03-14) and 53.0b2 build1
> (20170313154936) repeatedly, each time using a new profile.
> FX_MIGRATION_HISTORY_JANK_MS doesn't have quite constant values, even if
> running exactly the same steps every time. Here you have the values for each
> try
> https://docs.google.com/document/d/1Am9S0tMG1nptUx58dUeZe3jU8VbfXle-
> qmm4MHn_eAw/edit?usp=sharing. Also, if following the same steps but using
> the installed builds, the FX_MIGRATION_HISTORY_JANK_MS values seem to
> slightly increase. See
> https://docs.google.com/document/d/
> 10LpkRqN_Ns5YVmLFmwr2f1kPi1B8gbfopelrwZtlFfI/edit?usp=sharing.  

The numbers are influenced by any other things your machine is doing at the time (so if you have other browsers open with google docs, even what the google doc is doing could potentially influence this). The fluctuations seem within reason in these spreadsheets.
 
> > Do you see similar
> > numbers as in bug 1341097 in that case? If so, can you try to work out (by
> > trying the different combinations of items) what's causing the difference?
> Using latest aurora and beta builds, I didn't manage to reproduce similar
> FX_MIGRATION_HISTORY_JANK_MS values with those from
> https://bugzilla.mozilla.org/show_bug.cgi?id=1341097#c49, but only with
> 53.0b1 build1 (20170307064827). Those were encountered for complete
> migration and for bookmarks & history migration cases. See
> https://docs.google.com/document/d/
> 1R_00m69xUtGzaw6UC2vOKRzcfzfvpPl6ls9z8pIihyI/edit?usp=sharing. 
> 
> Please let me know if there is anything else to investigate.

OK. Thanks for all this detailed information. In this case I don't think there's much else to do here - we should just fix bug 1338800 and check if there's cases where we fire notifications for a pending import in a way that influences the subsequent import, or something like that.

Marking this 'fixed' as we finished investigating. Thanks again!