phoenix-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Andrew ...@starfishzone.com>
Subject Re: internal table client cache
Date Wed, 25 Jun 2014 20:02:58 GMT
+1

Another case which suffers the same issue:

client A runs a query / explain plan
client B drops an index
client A's queries / explain plans fail

This might sound like a strange use cases, but when you are testing 
performance and examining execution plans, I found myself using a SQL 
GUI and building big indexes in a different sqllline session.

What I'd like is either a connection string setting which aggressively 
refreshes the schema or a command like MySQL's FLUSH PRIVILEGES or SET 
xxx ON which gets intercepted by the driver.

Andrew.

On 25/06/2014 19:43, Jody Landreneau wrote:
> I have a use case where I have multiple instances of the phoenix 
> client running. Essentially, they are taking data and performing 
> upserts. These are on multiple machines. When I update a table schema, 
> ie)by adding a column, the clients start failing. The update is 
> performed via a client outside these running instances. The code that 
> builds the upsert statement understands that an additional column was 
> added and creates the proper upsert statement.
>
> Thinking this was a connection cache issue, I tried setting a max time 
> to close connections(they are in a pool). This did not work. I ended 
> up tracing the issue and finding that there is a MetaDataImpl cache 
> that gets populated on startup. Table schema are stored in this cache. 
> And when something is performed like an upsert there is code in the 
> FromCompiler that checks columns, but the columns are not updated in 
> this internal cache.
>
> I can file an issue on this with more details but wanted to get some 
> insight from others as to if it were some property to pass in that can 
> tell the cache to refresh at some interval. Possibly this should have 
> been stored on the connections and used that connection timeout or if 
> there was a property to pass in to the driver itself to cause a refresh.
>
> thanks --


Mime
View raw message