POC performance gain with a Email/list method

[chibenwa]

The idea: combines Email/query and Email/get in one go.

Why? Because it would do a single DB query do load all mails, instead of N.

I am mostly curious to know the possible performance gains with this approach.

I know we would remain 1 read per message for the preview and one other read per message for the headers... But theoretically this could be a 33% perf benefit for INBOX listing, which is quite common so worth a try...

Also it might allow reducing the JSON chatter a bit.

Spec

[ "Email/list", {
  "accountId": "rehielrbnev",
  "mailboxId": "ebverlner",
  "limit": 20,
  "offset": 0,
  "properties": [...]
}, "C0" ]

Definition of done

• Implement a minimalist version of Email/list
• Perf test it before / after, adapting gatling scripts as needed

[hungphan227]

My solution is adding extra columns to CassandraEmailQueryViewTable.

Currenlty, the table looks like:

CassandraModule MODULE = CassandraModule.
        .table(TABLE_NAME_RECEIVED_AT)
        .comment("Storing the JMAP projections for list of emails within a mailbox to not rely on OpenSearch for basic Email/query (sorts and filter on receivedAt).")
        .options(options -> options
            .withClusteringOrder(RECEIVED_AT, DESC)
            .withCaching(true, rows(DEFAULT_CACHED_ROW_PER_PARTITION)))
        .statement(statement -> types -> statement
            .withPartitionKey(MAILBOX_ID, DataTypes.UUID)
            .withClusteringColumn(RECEIVED_AT, DataTypes.TIMESTAMP)
            .withClusteringColumn(MESSAGE_ID, DataTypes.UUID)
            .withColumn(SENT_AT, DataTypes.TIMESTAMP))

I would add some extra fields so that there is no need to query from message table:

CassandraModule MODULE = CassandraModule.
        .table(TABLE_NAME_RECEIVED_AT)
        .comment("Storing the JMAP projections for list of emails within a mailbox to not rely on OpenSearch for basic Email/query (sorts and filter on receivedAt).")
        .options(options -> options
            .withClusteringOrder(RECEIVED_AT, DESC)
            .withCaching(true, rows(DEFAULT_CACHED_ROW_PER_PARTITION)))
        .statement(statement -> types -> statement
            .withPartitionKey(MAILBOX_ID, DataTypes.UUID)
            .withClusteringColumn(RECEIVED_AT, DataTypes.TIMESTAMP)
            .withClusteringColumn(MESSAGE_ID, DataTypes.UUID)
            .withColumn(SENT_AT, DataTypes.TIMESTAMP)
            .withColumn(PREVIEW, DataTypes.TEXT)
            .withColumn(HAS_ATTACHMENT, DataTypes.BOOLEAN)
            .withColumn(Flag.ANSWERED, BOOLEAN)
            .withColumn(Flag.DELETED, BOOLEAN)
            .withColumn(Flag.DRAFT, BOOLEAN)
            .withColumn(Flag.FLAGGED, BOOLEAN)
            .withColumn(Flag.RECENT, BOOLEAN)
            .withColumn(Flag.SEEN, BOOLEAN)
            .withColumn(Flag.USER, BOOLEAN)
            .withColumn(Flag.USER_FLAGS, setOf(TEXT))
            .withColumn(CassandraMessageV3Table.HEADER_CONTENT, TEXT))

[chibenwa]

@hungphan227 please do not take this task.

Pick another development task

My solution is adding extra columns to CassandraEmailQueryViewTable.

Pffff its obvious one should base this on messageidtable