Skip to content

Business intelligence made simple

Added cohorts for MySQL

Andrew Kane93dfbaf74fda
diff --git a/CHANGELOG.md b/CHANGELOG.md
index a313bf9e..0a6e4bd1 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,3 +1,7 @@
+## 2.4.1 (unreleased)
+
+- Added cohorts for MySQL
+
## 2.4.0 (2020-12-15)
- Added cohorts
diff --git a/lib/blazer/adapters/sql_adapter.rb b/lib/blazer/adapters/sql_adapter.rb
index f5bd73a1..3f0ca070 100644
--- a/lib/blazer/adapters/sql_adapter.rb
+++ b/lib/blazer/adapters/sql_adapter.rb
@@ -123,7 +123,7 @@ module Blazer
end
def supports_cohort_analysis?
- postgresql?
+ postgresql? || mysql?
end
# TODO treat date columns as already in time zone
@@ -131,8 +131,30 @@ module Blazer
raise "Cohort analysis not supported" unless supports_cohort_analysis?
cohort_column = statement =~ /\bcohort_time\b/ ? "cohort_time" : "conversion_time"
+ tzname = Blazer.time_zone.tzinfo.name
+
+ if mysql?
+ time_sql = "CONVERT_TZ(cohorts.cohort_time, '+00:00', ?)"
+ case period
+ when "day"
+ date_sql = "CAST(DATE_FORMAT(#{time_sql}, '%Y-%m-%d') AS DATE)"
+ date_params = [tzname]
+ when "week"
+ date_sql = "CAST(DATE_FORMAT(#{time_sql} - INTERVAL ((5 + DAYOFWEEK(#{time_sql})) % 7) DAY, '%Y-%m-%d') AS DATE)"
+ date_params = [tzname, tzname]
+ else
+ date_sql = "CAST(DATE_FORMAT(#{time_sql}, '%Y-%m-01') AS DATE)"
+ date_params = [tzname]
+ end
+ bucket_sql = "CAST(CEIL(TIMESTAMPDIFF(SECOND, cohorts.cohort_time, query.conversion_time) / ?) AS INTEGER)"
+ else
+ date_sql = "date_trunc(?, cohorts.cohort_time::timestamptz AT TIME ZONE ?)::date"
+ date_params = [period, tzname]
+ bucket_sql = "CEIL(EXTRACT(EPOCH FROM query.conversion_time - cohorts.cohort_time) / ?)::int"
+ end
# WITH not an optimization fence in Postgres 12+
+ # TODO improve performance for MySQL
statement = <<~SQL
WITH query AS (
#{statement}
@@ -143,14 +165,14 @@ module Blazer
GROUP BY 1
)
SELECT
- date_trunc(?, cohorts.cohort_time::timestamptz AT TIME ZONE ?)::date AS period,
+ #{date_sql} AS period,
0 AS bucket,
COUNT(DISTINCT cohorts.user_id)
FROM cohorts GROUP BY 1
UNION ALL
SELECT
- date_trunc(?, cohorts.cohort_time::timestamptz AT TIME ZONE ?)::date AS period,
- CEIL(EXTRACT(EPOCH FROM query.conversion_time - cohorts.cohort_time) / ?)::int AS bucket,
+ #{date_sql} AS period,
+ #{bucket_sql} AS bucket,
COUNT(DISTINCT query.user_id)
FROM cohorts INNER JOIN query ON query.user_id = cohorts.user_id
WHERE query.conversion_time IS NOT NULL
@@ -158,8 +180,7 @@ module Blazer
#{cohort_column == "conversion_time" ? "AND query.conversion_time != cohorts.cohort_time" : ""}
GROUP BY 1, 2
SQL
- tzname = Blazer.time_zone.tzinfo.name
- params = [statement, period, tzname, period, tzname, days.to_i * 86400]
+ params = [statement] + date_params + date_params + [days.to_i * 86400]
connection_model.send(:sanitize_sql_array, params)
end