Re: [HACKERS] Is anyone aware of data loss causing MultiXact bugs in 9.3.2?

2014-02-19 Thread Andres Freund
On 2014-02-18 18:10:02 -0800, Peter Geoghegan wrote: On Tue, Feb 18, 2014 at 5:50 PM, Alvaro Herrera alvhe...@2ndquadrant.com wrote: Peter Geoghegan wrote: I've had multiple complaints of apparent data loss on 9.3.2 customer databases. There are 2 total, both complaints from the past week,

Re: [HACKERS] Is anyone aware of data loss causing MultiXact bugs in 9.3.2?

2014-02-19 Thread Peter Geoghegan
On Wed, Feb 19, 2014 at 12:40 AM, Andres Freund and...@2ndquadrant.com wrote: Was there an index only scan or just a index scan? Any chance of a corrupted index? Just an index scan. I think it's unlikely to be a corrupt index, because the customer said that he dropped and restored the index,

Re: [HACKERS] Is anyone aware of data loss causing MultiXact bugs in 9.3.2?

2014-02-19 Thread Andres Freund
On 2014-02-19 00:55:03 -0800, Peter Geoghegan wrote: On Wed, Feb 19, 2014 at 12:40 AM, Andres Freund and...@2ndquadrant.com wrote: Was there an index only scan or just a index scan? Any chance of a corrupted index? Just an index scan. I think it's unlikely to be a corrupt index, because

[HACKERS] Is anyone aware of data loss causing MultiXact bugs in 9.3.2?

2014-02-18 Thread Peter Geoghegan
I've had multiple complaints of apparent data loss on 9.3.2 customer databases. There are 2 total, both complaints from the past week, one of which I was able to confirm. The customer's complaint is that certain rows are either visible or invisible, depending on whether an index scan is used or a

Re: [HACKERS] Is anyone aware of data loss causing MultiXact bugs in 9.3.2?

2014-02-18 Thread Alvaro Herrera
Peter Geoghegan wrote: I've had multiple complaints of apparent data loss on 9.3.2 customer databases. There are 2 total, both complaints from the past week, one of which I was able to confirm. The customer's complaint is that certain rows are either visible or invisible, depending on whether

Re: [HACKERS] Is anyone aware of data loss causing MultiXact bugs in 9.3.2?

2014-02-18 Thread Peter Geoghegan
On Tue, Feb 18, 2014 at 5:50 PM, Alvaro Herrera alvhe...@2ndquadrant.com wrote: The multixact bugs would cause tuples to be hidden at the heap level. If the tuples are visible in a seqscan, then these are more likely to be related to index problems, not multixact problem. That was my first

Re: [HACKERS] Is anyone aware of data loss causing MultiXact bugs in 9.3.2?

2014-02-18 Thread Peter Geoghegan
On Tue, Feb 18, 2014 at 5:56 PM, Peter Geoghegan p...@heroku.com wrote: That was my first suspicion, but then re-indexing didn't help, while VACUUM FREEZE had the immediate effect of making both plans give a consistent answer. I should add that this is an unremarkable int4 primary key (which

Re: [HACKERS] Is anyone aware of data loss causing MultiXact bugs in 9.3.2?

2014-02-18 Thread Peter Geoghegan
On Tue, Feb 18, 2014 at 5:50 PM, Alvaro Herrera alvhe...@2ndquadrant.com wrote: Peter Geoghegan wrote: I've had multiple complaints of apparent data loss on 9.3.2 customer databases. There are 2 total, both complaints from the past week, one of which I was able to confirm. The customer's