通过 DataMapper 来自 SQLite 内存数据库的 No Such Table 错误

我有一个 Ruby 程序，它使用 DataMapper 作为 ORM 与内存中的 SQLite DB 进行通信。这一直工作得很好，但是我最近刚刚添加了一个新的 DM 类和相应的表。令我惊讶的是，现在在 auto_migrate 过程中事情发生了爆炸！

这是 DataMapper 生成的 SQL：

~ (0.000390) PRAGMA table_info("sensationd_channels")
~ (0.000010) PRAGMA table_info("sensationd_commands")
~ (0.000009) PRAGMA table_info("sensationd_configurations")
~ (0.000052) PRAGMA table_info("sensationd_measurements")
~ (0.000028) SELECT sqlite_version(*)
~ (0.000035) DROP TABLE IF EXISTS "sensationd_channels"
~ (0.000009) PRAGMA table_info("sensationd_channels")
~ (0.000423) CREATE TABLE "sensationd_channels" ("id" INTEGER NOT NULL PRIMARY KEY AUTOINCREMENT, "channel" INTEGER NOT NULL, "name" VARCHAR(50), "precision" INTEGER DEFAULT 11, "gain" INTEGER DEFAULT 1, "differential" BOOLEAN DEFAULT 'f', "configuration_id" INTEGER NOT NULL)
~ (0.000191) CREATE INDEX "index_sensationd_channels_configuration" ON "sensationd_channels" ("configuration_id")
~ (0.000015) DROP TABLE IF EXISTS "sensationd_commands"
~ (0.000009) PRAGMA table_info("sensationd_commands")
~ (0.000153) CREATE TABLE "sensationd_commands" ("id" INTEGER NOT NULL PRIMARY KEY AUTOINCREMENT, "action" INTEGER DEFAULT 1, "complete" BOOLEAN DEFAULT 'f', "issued_at" TIMESTAMP, "completed_at" TIMESTAMP)
~ (0.000015) DROP TABLE IF EXISTS "sensationd_configurations"
~ (0.000009) PRAGMA table_info("sensationd_configurations")
~ (0.000155) CREATE TABLE "sensationd_configurations" ("id" INTEGER NOT NULL PRIMARY KEY AUTOINCREMENT, "created_on" TIMESTAMP, "modified_on" TIMESTAMP, "name" VARCHAR(50) NOT NULL, "active" BOOLEAN)
~ (0.000015) DROP TABLE IF EXISTS "sensationd_measurements"
~ (0.000009) PRAGMA table_info("sensationd_measurements")
~ (0.000152) CREATE TABLE "sensationd_measurements" ("id" INTEGER NOT NULL PRIMARY KEY AUTOINCREMENT, "timestamp" TIMESTAMP, "measurement" VARCHAR(65535) NOT NULL, "channel_id" INTEGER NOT NULL, "channel_configuration_id" INTEGER NOT NULL)
~ (0.000175) CREATE INDEX "index_sensationd_measurements_channel" ON "sensationd_measurements" ("channel_id", "channel_configuration_id")
~ (0.000083) SELECT "id", "created_on", "modified_on", "name", "active" FROM "sensationd_configurations" WHERE "active" = 't' ORDER BY "id" LIMIT 1
~ (0.000073) INSERT INTO "sensationd_configurations" ("created_on", "modified_on", "name", "active") VALUES ('2011-08-01T12:36:18-07:00', '2011-08-01T12:36:18-07:00', 'Test U6-Pro Configuration, differential.', 't')
~ (0.000109) SELECT "id", "action", "complete", "issued_at", "completed_at" FROM "sensationd_commands" ORDER BY "issued_at" DESC LIMIT 1
~ (0.000086) INSERT INTO "sensationd_channels" ("channel", "name", "precision", "gain", "differential", "configuration_id") VALUES (0, '0', 11, 0, 't', 1)
~ no such table: sensationd_commands (code: 1, sql state: , query: SELECT "id", "action", "complete", "issued_at", "completed_at" FROM "sensationd_commands" ORDER BY "issued_at" DESC LIMIT 1, uri: sqlite3::memory:?scheme=sqlite&user=&password=&host=&port=&query=&fragment=&adapter=sqlite3&path=:memory:)

看起来它创建表的过程很顺利，但是几行之后就找不到它了。我认为我错误配置了数据库连接，除了找到其他表并且工作正常之外。

正在使用的软件：

• 通过 RVM 的 Ruby 1.9.2p289
• SQLite3 @3.7.7.1 通过 MacPorts
• DataMapper gem v 1.1.0

有谁知道为什么会出现这种情况，我能做些什么？

我怀疑，问题是由于 DataMapper（或更准确地说，DataObjects，数据库驱动程序 DataMapper 使用）自动执行的线程池所致。数据库连接不在线程之间共享。对于像 postgresql 或 mysql 甚至 sqlite3 这样的“文件支持”数据库来说，这很好（甚至是有益的）。对于 sqlite3 的内存存储，连接is数据库。因此，其他线程将因这个原因而失败。此外，经过一段时间的不活动（大约 1 分钟？）后，线程将被清理，数据库也将消失。

如果是这样，我不确定是否有简单的解决方法。您也许可以修改 do_sqlite3 来避免这种情况。另一种基本上应该同样快的替代方案是在 ramdrive 上使用文件支持的 sqlite3 DB。