0

I have created a materialized view using this query

CREATE MATERIALIZED VIEW db.top_ids_mv (
`date_time` DateTime,
`id` String,
`total` UInt64
) ENGINE = SummingMergeTree
  ORDER BY
  (date_time, id) SETTINGS index_granularity = 8192 POPULATE AS
SELECT
  toDateTime((intDiv(toUInt32(date_time), 60 * 60) * 60) * 60) AS date_time,
  id AS id,
  count(*) AS count
FROM
  db.table
WHERE
  type = 'user'
GROUP BY
  date_time,id

My table contains almost 18 billion records. I have inserted my old data using POPULATE. But newly inserted data is not getting inserted into this materialized view. I have created many other views and they are working fine but this is creating issue.

This is what I am receiving in logs

2021.09.23 19:54:54.424457 [ 137949 ] {5b0f3c32-2900-4ce4-996d-b9696bd38568} <Trace> PushingToViewsBlockOutputStream: Pushing (sequentially) from db.table (15229c91-c202-4809-9522-9c91c2028809) to db.top_ids_mv (0cedb783-bf17-42eb-8ced-b783bf1742eb) took 0 ms.

One thing I noticed is that it is taking 0ms. I think that is wrong because query must take some time.

Thanks. Any help would be appreciated

Zunnurain Badar
  • 920
  • 2
  • 7
  • 28

1 Answers1

1
  1. SummingMergeTree does not store rows with metrics == 0.
  2. total UInt64 <----> count(*) AS count -- names does not match. Your Mat.View inserts 0 into total, count goes nowhere.

Both are expected and specifically implemented for the reasons.

https://den-crane.github.io/Everything_you_should_know_about_materialized_views_commented.pdf

...
SELECT
  toDateTime((intDiv(toUInt32(date_time), 60 * 60) * 60) * 60) AS date_time,
  id AS id,
  count(*) AS total       --<<<<<------
FROM
  db.table
...

For query performance and better data compression I would do

ENGINE = SummingMergeTree
  ORDER BY   ( id, date_time ) --- order id , time

Also try codecs

`date_time` DateTime CODEC(Delta, LZ4),
`id` LowCardinality(String),
`total` UInt64 CODEC(T64, LZ4)
Denny Crane
  • 11,574
  • 2
  • 19
  • 30