Bug 229646 – Migrate PrivateClickMeasurements from ResourceLoadStatistics database to new database

• Component: New Bugs
• Status: RESOLVED FIXED

[Alex Christensen]

Migrate PrivateClickMeasurements from ResourceLoadStatistics database to new database

[Alex Christensen]

Created attachment 436713 [details]
Patch

[Alex Christensen]

Created attachment 436722 [details]
Patch

[Kate Cheney]

Comment on attachment 436722 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=436722&action=review

This is fine but I think we can remove a lot of migration code now that we convert the database entries into PCM objects before re-inserting. We shouldn't need to rename columns or even add new ones I don't think.

> Source/WebKit/NetworkProcess/Classifier/ResourceLoadStatisticsDatabaseStore.cpp:537
> +        constexpr auto allAttributedPrivateClickMeasurementQuery = "SELECT *, MIN(earliestTimeToSendToSource, earliestTimeToSendToDestination) as minVal "

We sort these later in PrivateClickMeasurementDatabase when we pass them to PrivateClickMeasurementManager, so no need to sort them here. You can just use SELECT * FROM AttributedPrivateClickMeasurement.

> Source/WebKit/NetworkProcess/Classifier/ResourceLoadStatisticsDatabaseStore.cpp:639
> +void ResourceLoadStatisticsDatabaseStore::renameColumnInTable(String&& tableName, String&& existingColumnName, String&& expectedColumnName)

You shouldn't need this anymore if you use SELECT * statements when migrating to the new schema, because they don't rely on the name of the columns to migrate the data.

> Source/WebKit/NetworkProcess/Classifier/ResourceLoadStatisticsDatabaseStore.cpp:654
> +void ResourceLoadStatisticsDatabaseStore::addMissingColumnsToTable(String&& tableName, const Vector<String>& existingColumns, const Vector<String>& expectedColumns)

You shouldn't need this either. I think SQLite returns null values if the requested column doesn't exist, so the call to buildPrivateClickMeasurementFromDatabase() will return null values for keyID, token, signature, and earliestTimeToSend even if the columns don't exist. It might be nice to be safe and check for null strings before calling attribution.setSourceSecretToken() so we don't create a token with only null values.

> Source/WebKit/NetworkProcess/Classifier/ResourceLoadStatisticsDatabaseStore.cpp:679
> +void ResourceLoadStatisticsDatabaseStore::addMissingColumnsIfNecessary()

Ditto, you can remove this.

> Source/WebKit/NetworkProcess/Classifier/ResourceLoadStatisticsDatabaseStore.cpp:735
> +void ResourceLoadStatisticsDatabaseStore::renameColumnsIfNecessary()

Ditto, I think we can remove.

> Source/WebKit/NetworkProcess/PrivateClickMeasurement/PrivateClickMeasurementDatabase.cpp:153
> +        ASSERT(sourceEarliestTimeToSend != -1);

I think we should adjust this to be ASSERT(sourceEarliestTimeToSend != -1 || destinationEarliestTimeToSend != -1);

It is possible you will migrate an entry that was already sent to the source but not the destination, resulting in a sourceEarliestTimeToSend value of -1.

> Tools/TestWebKitAPI/Tests/WebKitCocoa/PrivateClickMeasurement.mm:58
> +void addValuesToTable(WebCore::SQLiteDatabase& database, ASCIILiteral query, std::array<Variant<StringView, int, double>, size> values)

fancy tests!

> Tools/TestWebKitAPI/Tests/WebKitCocoa/PrivateClickMeasurement.mm:88
> +        "earliestTimeToSend REAL, "

To mimic the actual v1, earliestTimeToSend should be REAL NOT NULL.

> Tools/TestWebKitAPI/Tests/WebKitCocoa/PrivateClickMeasurement.mm:117
> +        "earliestTimeToSend REAL, token TEXT, signature TEXT, keyID TEXT, "

Ditto, this should be REAL NOT NULL for an accurate test.

[Alex Christensen]

Created attachment 436814 [details]
Patch

[Kate Cheney]

Comment on attachment 436814 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=436814&action=review

r=me

> Source/WebKit/NetworkProcess/Classifier/ResourceLoadStatisticsDatabaseStore.h:261
> +    mutable std::unique_ptr<WebCore::SQLiteStatement> m_checkIfTableExistsStatement;

I don't think this needs to be mutable.

> Source/WebKit/NetworkProcess/PrivateClickMeasurement/PrivateClickMeasurementDatabase.h:92
> +    mutable Statement m_insertObservedDomainStatement;

Do these all need to be mutable? I don't think most are used in const functions.

[Alex Christensen]

Comment on attachment 436814 [details]
Patch

They're all conceptually caches, which is what "mutable" is for.  Making them mutable allows them to be used in const contexts.  I needed to add some const, so I made them all mutable so that we don't need to hurt future change history by adding it.

[EWS]

Committed r281779 (241116@main): <https://commits.webkit.org/241116@main>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 436814 [details].

[Radar WebKit Bug Importer]

<rdar://problem/82550832>