sqlite3_stmt* stmt2 = NULL;
executeStmt(db, stmt, "insert into foo(id) values(10)");
sqlite3_stmt* stmt3 = NULL;
executeStmt(db, stmt3, "update foo set content=zeroblob(10*1024*1024) where rowid=10");
sqlite3_stmt* stmt4 = NULL; executeStmt(db, stmt4, "update foo set content2=zeroblob(1024) where rowid=10"); // OOM here
不过是0单列的0填充而已,怎么会OOM呢?
我直接给它的作者D. Richard Hipp和他的team发邮件询问,得到的答复是:
每次更新,sqlite都会读取该行的数据,加载到内存里。难怪,第一列可是要10M内存。
Richard建议:使用incremental BLOB I/O mechanism机制进行读写。
可是这个机制局限性依然明显:
e.g.
first write, 20K, then append 10K. If we want to append it, we need to zeroblob 30K first. Otherwise, sqlite_blob_write will fail to write the chunk.
???
他的答复是:
Unfortunately, no; there is no way to increase the size of a row without allocating memory sufficient to hold the entire row.
除非要更改sqlite的文件格式。一旦更改,将会打破与以前sqlite文件的兼容性。那是不可容忍的。