Concurrent insert operations causing deadlock(并发插入时, 触发死锁)

[Sube-py]

I am using jieba as the tokenizer and have configured triggers as per the documentation. During batch data insertion in my project, deadlocks are triggered. Additional questions:

• Does index creation lock the table?
• Are index statistics (e.g., document frequency) calculated per table or per row? My table contains multiple documents.

我使用了jieba作为分词器, 并按照文档设置了触发器, 当我在项目中批量插入数据时, 会触发死锁, 同时我有几个问题:

1. 索引建立时会锁表吗?
2. 索引统计的文档频率是整表的还是整行的? 我的表里会包含多篇文档

[VoVAllen]

Sorry for the late reply. The document frequency is across all the row, so it will handle multiple documents by the index. The index will collect the frequency info for all the documents

[VoVAllen]

For deadlock, could you provide more information about it ?

[Sube-py]

For deadlock, could you provide more information about it ?

Thks for reply, I will provide a demo this weekend.

[Sube-py]

oh, pg_tokenizer.rs caused this bug

2025-07-04 02:54:06.227 UTC [475] ERROR:  canceling statement due to user request
2025-07-04 02:54:06.227 UTC [475] CONTEXT:  while inserting index tuple (154,182) in relation "model_jieba_model"
	SQL statement "
	    WITH 
	    new_tokens AS (
	        SELECT unnest(tokenizer_catalog.apply_text_analyzer_for_custom_model($1.text, 'jieba_analyzer')) AS token
	    ),
	    to_insert AS (
	        SELECT token FROM new_tokens
	        WHERE NOT EXISTS (
	            SELECT 1 FROM tokenizer_catalog."model_jieba_model" WHERE token = new_tokens.token
	        )
	    )
	    INSERT INTO tokenizer_catalog."model_jieba_model" (token) SELECT token FROM to_insert ON CONFLICT (token) DO NOTHING"
	PL/pgSQL function custom_model_insert_trigger() line 7 at EXECUTE