用R的pool包重新连接到PostgreSQL数据库

Question

我有一个用R plumber构建的API ，它使用RPostgreSQL和pool连接到PostgreSQL数据库（尽管如果我使用的是Shiny应用程序，这也适用）：

# create the connection pool
pool <- dbPool(
  drv = PostgreSQL(),
  host = Sys.getenv("DB_HOST"),
  port = 5432,
  dbname = "db",
  user = Sys.getenv("DB_USER"),
  password = Sys.getenv("DB_PASSWORD")
)

# start the API
pr <- plumb("plumber.R")

# on stop, close the pool
pr$registerHooks(
  list("exit" = function() { poolClose(pool) })
)

我想每天导入新数据。最简单的方法是创建一个新数据库并将其推广到生产环境：

CREATE DATABASE db_new;
-- create the tables
-- bulk-insert the data
SELECT pg_terminate_backend (pid) FROM pg_stat_activity WHERE datname = 'db';
DROP DATABASE db;
ALTER DATABASE db_new RENAME TO db;

这是快速的，最大程度地减少了停机时间。问题是，pool然后丢失了与数据库的连接，并且不会自动尝试重新连接：

> tbl(pool, "users")
Error in postgresqlExecStatement(conn, statement, ...) : 
  RS-DBI driver: (could not Retrieve the result : FATAL:  terminating connection due to administrator command
server closed the connection unexpectedly
    This probably means the server terminated abnormally
    before or while processing the request.
)

即使我不是每天更换数据库，数据库服务器也会偶尔重启，这也会导致我的应用程序崩溃。重新连接似乎不是池，RPostgreSQL或DBI的功能。有谁知道解决这个问题的方法？

Answer 1

我最近遇到了一个类似的问题，因为当实例wait_timeout超出时关闭 MySQL 连接。我在 RStudio Community 上看到了您的帖子，并受到了您的解决方案的启发。如果您仍在使用它，并且正在寻找在包装您使用的实际函数时避免额外查询的解决方案，这里有一个 reprex 演示我想出的东西，以及一个证明它有效的例子：

library(dplyr, warn.conflicts = FALSE)
library(pool)
library(RMariaDB)

generate_safe_query <- function(pool) {
  function(db_function, ...) {
    tryCatch({
      db_function(pool, ...)
    }, error = function(e) {
      if (grepl("Lost connection to MySQL server during query", e$message)) {
        # Preserve `validationInterval` so that it can be restored
        validation_interval <- pool$validationInterval
        # Trigger destruction of dead connection
        pool$validationInterval <- 0
        refreshed_connection <- poolCheckout(pool)
        poolReturn(refreshed_connection)
        # Restore original `validationInterval`
        pool$validationInterval <- validation_interval
        # Execute the query with the new connection
        db_function(pool, ...)
      } else {
        # Unexpected error
        stop(e)
      }
    })
  }
}

mysql_pool <- dbPool(MariaDB(),
                     host = "127.0.0.1",
                     username = "root",
                     password = "",
                     dbname = "test")

safe_query <- generate_safe_query(mysql_pool)

# Works
safe_query(tbl, "notes")
#> # Source:   table<notes> [?? x 2]
#> # Database: mysql 8.0.15 [root@127.0.0.1:/test]
#>      id note 
#>   <int> <chr>
#> 1     1 NOTE1

# Set the `wait_timeout` to 5 seconds for this session
invisible(safe_query(dbExecute, "SET SESSION wait_timeout = 5"))

# Wait longer than `wait_timeout` to trigger a disconnect
Sys.sleep(6)

# Still works; warning will appear notifying that connection was
# destroyed and replaced with a new one
safe_query(tbl, "notes")
#> Warning: It wasn't possible to activate and/or validate the object. Trying
#> again with a new object.
#> # Source:   table<notes> [?? x 2]
#> # Database: mysql 8.0.15 [root@127.0.0.1:/test]
#>      id note 
#>   <int> <chr>
#> 1     1 NOTE1

safe_query(poolClose)
# Or, equivalently: 
# poolClose(mysql_pool)

^{由reprex 包(v0.3.0)于 2019 年 5 月 30 日创建}

该函数返回通过generate_safe_query将工作与任何数据库查询功能（例如dbExecute，dbGetQuery等）。显然，您需要更新它匹配的错误消息以满足您的需要。

我还打开了我自己的社区主题，我认为应该包含一个选项dbPool，以减少对此类变通方法的需求。