Hikaricp: Illegal argument for PostgreSQL JDBC setNetworkTimeout

[MathiasBerwig]

Environment

HikariCP version: 2.7.6
JDK version     : 1.8.0_111
Database        : PostgreSQL
Driver version  : 42.2.1
G Cloud SQL Socket Factory version: 1.0.5

---

I'm using HickariCP with Google App Engine Standard and Google Cloud SQL Socket Factory to deploy an Java 8 API and facing a PSQLException that is thrown most of the times I upload the app to GAE.

Some important points:

• The updates with support for the socket factory are just a few days old and PGSQL is beta on GAE AFAIK, so I'm not 100% sure if the problem is with HikariCP or the driver or the socket factory;
• When connecting to database as a external application (without Google Cloud Socket Factory), everything works just fine (except the fact I need to allow any IP address to connect to my DB instance - each new instance seems to have a different IP, so I need to white list all);
• The PSQLException if refer is thrown at this line and indicates that an negative value was passed as argument to the PgConnection.setNetworkTimeout;
• If I set the connection timeout on code or Properties, the error occurs the same way.
• After several tries I'm able to make it work, just by re-submitting the code again and again.
• I've tried warming up the instance, deleting and recreating services, updating existing versions and creating new ones.

Here is the log:

javax.servlet.ServletContext log: RestletServlet: [Restlet] Attaching application: br.com.appkizumba.api.WebApiApplication@4dd025df to URI: /v1
org.restlet.Application start: Starting br.com.appkizumba.api.WebApiApplication application (Application.java:589)
INFO com.zaxxer.hikari.HikariDataSource - HikariPool-1 - Starting...
com.google.cloud.sql.postgres.SocketFactory createSocket: Connecting to Cloud SQL instance [...]. (SocketFactory.java:55)
com.google.cloud.sql.postgres.SocketFactory createSocket: GAE Unix Sockets (SocketFactory.java:63)
[Request11B37C49] ERROR com.zaxxer.hikari.pool.HikariPool - HikariPool-1 - Exception during pool initialization.
org.postgresql.util.PSQLException: Network timeout must be a value greater than or equal to 0.
    at org.postgresql.jdbc.PgConnection.setNetworkTimeout(PgConnection.java:1540)
    at com.zaxxer.hikari.pool.PoolBase.setNetworkTimeout(PoolBase.java:551)
    at com.zaxxer.hikari.pool.PoolBase.setupConnection(PoolBase.java:438)

[brettwooldridge]

@MathiasBerwig To me this looks like an issue with GAE, but possibly a driver issue. The exception above indicates that the driver returned a negative number as the current network timeout setting. When HikariCP tries to restore the default, after changing it temporarily, the driver complains.

The question is why did the driver return a negative number as the current timeout? The default should be 0 (zero), which indicates infinite timeout...

EDIT: It should be noted that this issue is not seen with the regular driver, which makes me suspect the GAE SocketFactory. I would report the issue to Google. The network timeout returned from Connection.getNetworkTimeout() should be zero (0) by default.

[maricn]

@MathiasBerwig +1
I'm experiencing the same issue right now, more or less same versions. Could you, just for the reference share your findings and if/how you resolved this issue? Thank you.

[MathiasBerwig]

@maricn I did not find a reasonably secure way to fix this issue. Due to the startup-like nature of my project, I had enabled external connections and am using the regular driver and socket factory to connect to Cloud SQL by IP.

Please note that each new GAE instance will have a new and probably unknown IP address, so you'd need to whitelist any address (0.0.0.0/0). This is a huge security flaw (bots are trying to guess the password all day).

[brettwooldridge]

@MathiasBerwig @maricn I can investigate a "fix" (workaround) to ignore the network timeout if the get call returns a negative number. However, my strong preference would be that either or both of you two open an issue/incident report with GAE. Even if they don't support network timeout (which they should), if they just return a zero (0) value then all will be right in the world.

EDIT: If the GAE issue has a linkable URL, please post it here for future readers.

[maricn]

@MathiasBerwig
Thank you for your prompt response. I understand, we're in the similar position, also considering switching to GAE flex, since whitelisting everything is a no-go here.

Just for the reference, using some [or none] connection pool (eg. org.postgresql.ds.PGSimpleDataSource) circumvents the issue.

@brettwooldridge
I'm right on your side. I've created an issue with GAE

[brettwooldridge]

@maricn Thanks. I "starred" that issue. @MathiasBerwig I encourage you to do so as well.

[MathiasBerwig]

I'm no longer using HikariCP as my project is no more active. Since nobody else seems to be facing this problem, I'm closing the issue.