OpenSAF / Tickets / #1467 cpav:two apps on same node simultaneously failed to write checkpoint and app hangs

Bug Analysis :

checkpointHandle - is A pointer to the checkpoint handle, allocated in the
address space of the invoking process (CPA/application) . CPA stores into the
memory area of CPA/application/process uses to access the checkpoint in
subsequent invocations of the functions of the Checkpoint Service API.
In the case of saCkptCheckpointOpenAsync() , saCkptCheckpointWrite() ,
this handle is returned in the corresponding response message.

Eevn though saCkptCheckpointWrite() is Sync request checkpointHandle is used by CPND for tracking
CPND<--->CPND messaging invoking activity on request of saCkptCheckpointWrite() .

If the ckpoint is SA_CKPT_CHECKPOINT_COLLOCATED & SA_CKPT_WR_ALL_REPLICAS checkpoint ,
and the checkpoint is opened on multiple nodes , and saCkptCheckpointWrite() are beeing requested by
multiple CPAs from same Node , then the local CPND to update the all other CPNDS ,
whic has opened the same Ckpt, to track the pending invocations/event a
teperory cpnd_evt entry added with key checkpointHandle and after a successful write responserevived
from all the CPND's , using this cpnd_evt entry the local CPND will response back to CPA/application with result,
and then deleted cpnd_evt entry from local CPND ( temporary only for tracking response message of peer CPND )

In current code CPA is using malloc() return value as checkpointHandle for
unique reference key where malloc() returns virtual memory specific to processes,
so malloc() can return the same pointer value in separate CPA processes .

If we are running two ckpt app (2 processes) try to write data into two
difference/same checkpoints , it is possible that the checkpointHandle is being passed as
same from both checkpoint application processes , as same checkpointHandle is being shared to CPND by both CPA
as key reference, the CPND can miss behave because of ambiguous same reference key.

Solution :

As malloc() is standred call and the Checkpoint Service API Specification says
checkpointHandle should be pointer allocated in the address space of the invoking
process/CPA , CPND will return try again before the saCkptCheckpointWrite() call ,
if multiple saCkptCheckpointWrite() request come to CPND with SAME
checkpointHandle ( same virtual memory address) from different CPA's at the same time.

Patch will be published soon.

changeset: 6802:a9b393bb1c66
tag: tip
parent: 6799:93e72338e78f
user: A V Mahesh mahesh.valla@oracle.com
date: Mon Sep 14 12:14:30 2015 +0530
summary: cpsv: prevented multiple write request with same checkpointHandle at cpnd [#1467]

changeset: 6801:bd1ac2c47a02
branch: opensaf-4.6.x
parent: 6798:de765110343d
user: A V Mahesh mahesh.valla@oracle.com
date: Mon Sep 14 12:13:57 2015 +0530
summary: cpsv: prevented multiple write request with same checkpointHandle at cpnd [#1467]

changeset: 6800:47e4745bd82a
branch: opensaf-4.5.x
parent: 6797:84dee6e50f1e
user: A V Mahesh mahesh.valla@oracle.com
date: Mon Sep 14 12:12:59 2015 +0530
summary: cpsv: prevented multiple write request with same checkpointHandle at cpnd [#1467]

Tickets: ~~#1467~~

cpav:two apps on same node simultaneously failed to write checkpoint and app hangs

Milestone

Searches

Help

#1467 cpav:two apps on same node simultaneously failed to write checkpoint and app hangs

Steps to reproduce :

Related

Discussion

Bug Analysis :

Solution :

Related