Yii2 ActiveRecord identity map memory growth in long-running CLI batch jobs

Answer first

Yes. Yii2 ActiveRecord keeps one object instance per primary key in a process-wide static cache. In a long-running console command that iterates thousands of rows, that cache grows monotonically unless you clear it. An array-based DbCache provider also holds query results in RAM for the lifetime of the process, compounding growth.

Confirmed facts

• ActiveRecord stores loaded instances in ActiveRecord::$instanceCache. It is a static array that lives for the entire PHP process.
• From Yii 2.0.27 onward the framework exposes ActiveRecord::$enableIdentityMap to turn the map on/off globally, and ActiveRecord::clearInstanceCache() to empty the static array without touching the database.
• DbCache with an in-memory provider such as ArrayCache stores cached results in PHP memory for the process lifetime. The cache is not evicted between batches.
• asArray() or a plain DbCommand bypasses ActiveRecord objects entirely, so no identity-map entries are created for that query.

Likely explanation vs confirmed

Likely explanation: steady memory growth in your console command is dominated by the identity map retaining AR objects, and possibly by ArrayCache holding large result sets. This is consistent with Yii2 design, but the dominant source in your specific job can only be confirmed by measuring memory_get_usage() before/after clearInstanceCache() and inspecting cache size.

Supported ways to bound or release the map

Bound globally

Set before the batch starts, in Yii 2.0.27+:

\yii\db\ActiveRecord::$enableIdentityMap = false;

This disables identity-map population for the whole process. It can break code that relies on AR identity, e.g., relation caching and findOne() returning the same instance within the same request. Use only when the batch does not depend on identity semantics.

Release between batches without resetting schema

Keep identity map enabled but clear it per chunk:

$query = Model::find()->where([...])->orderBy('id');
foreach ($query->batch(1000) as $rows) {
    foreach ($rows as $model) {
        // process
    }
    \yii\db\ActiveRecord::clearInstanceCache();
    // optional: gc_collect_cycles();
}

clearInstanceCache() empties the static instance cache only. It does not reset DB schema, connections, or other caches.

Avoid identity map for read-only scans

• Use Model::find()->asArray()->batch(1000) to get arrays instead of objects.
• Use a DbCommand for pure projection.

DbCache with array-based provider

Yes. With ArrayCache the cache lives in PHP memory for the lifetime of the persistent CLI process. Repeated DbCache queries will keep the serialized result sets in RAM until the process ends or you manually $cache->flush(). For long-running CLI jobs avoid ArrayCache for query caching; use a file or Redis provider, or disable DbCache for the batch.

Minimal verification

1. Log peak memory before and after a chunk, then call clearInstanceCache() and log again.
2. Inspect count(\yii\db\ActiveRecord::$instanceCache) before/after clearing.
3. With DbCache enabled, run the same cached query twice and check memory does not drop on the second run.

One missing detail that changes the recommendation: which Yii 2.x version is in use. $enableIdentityMap and clearInstanceCache() are documented from 2.0.27 onward; earlier releases lack these APIs and require the asArray()/DbCommand workaround.