Ok. Thanks. But I have another question about checkpoint iteration, for current implementation, if the section that the current iterator points to is delete, no further section will be got? Is there a fix for this issue?
Ted From: A V Mahesh [mailto:[email protected]] Sent: Tuesday, May 26, 2015 4:21 PM To: Yao Cheng LIANG; [email protected] Subject: Re: [users] ckptnd memory leak(very serious) Hi, The current implementations is matching , based on saCkptCheckpointTrackCallback specification it is not not about simply customizing CPA code. -AVM On 5/26/2015 12:31 PM, Yao Cheng LIANG wrote: This is because the following code in cpa_proc_ckpt_arrival_ntfy: callback->ioVector[i].dataBuffer = NULL; callback->ioVector[i].readSize = 0; callback->ioVector[i].dataSize = ckpt_data->dataSize; callback->ioVector[i].dataOffset = ckpt_data->dataOffset; if I revise the first two lines like this: callback->ioVector[i].dataBuffer = ckpt_data->data; callback->ioVector[i].readSize = readSize; Then agent will get the data, right? Thanks. Ted From: A V Mahesh [mailto:[email protected]] Sent: Tuesday, May 26, 2015 2:55 PM To: Yao Cheng LIANG; [email protected]<mailto:[email protected]> Subject: Re: [users] ckptnd memory leak(very serious) On 5/26/2015 12:06 PM, Yao Cheng LIANG wrote: can I pass in read_size no zero, and then the updated data will be sent to local agent? NO , you will always get get zero size , irreverent of read_size , because In CPA it is made to zero so application will get zero . -AVM On 5/26/2015 12:06 PM, Yao Cheng LIANG wrote: Dear Mahesh, Thanks for these information. But If I do not want these strict synchronization, can I pass in read_size no zero, and then the updated data will be sent to local agent? If I want local agent to read data from CPA_EVT_ND2A_CKPT_ARRIVAL_NTFY, I have to revise agent part of ARRIVAL_NTFY, now it supposes passed in data is NULL and read-size also zero. Right? Ted From: A V Mahesh [mailto:[email protected]] Sent: Tuesday, May 26, 2015 2:29 PM To: Yao Cheng LIANG; [email protected]<mailto:[email protected]> Subject: Re: [users] ckptnd memory leak(very serious) Hi Ted, On 5/26/2015 11:49 AM, Yao Cheng LIANG wrote: But for every write, the non-active cpnd will get updated with the fresh data, it should not be any problem if it update its local agents. Right? Write Lock is at CPND level , and in some ckpts the active replica can be diffrent node , taht may allow some space to another agent to write while another agent received data in track callback ( it can be older data some other data is simultaneously written in to same offset) so it always preferred to read the data from the replica. -AVM Ted From: A V Mahesh [mailto:[email protected]] Sent: Tuesday, May 26, 2015 2:14 PM To: Yao Cheng LIANG; [email protected]<mailto:[email protected]> Subject: Re: [users] ckptnd memory leak(very serious) HI Ted, On 5/26/2015 11:18 AM, Yao Cheng LIANG wrote: Do you think it is possible to pass the data along with the section to agents, it will be more efficient This approach will introduce out-of-order data read by application , because of concurrent/continuous write in to a checkpoint . -AVM On 5/26/2015 11:18 AM, Yao Cheng LIANG wrote: Dear Mahesh, When cpsv agent update the sections, the request will be sent to active replica, after apply these changes, cpns where avtive replica reside will call cpnd_ckpt_update_replica to update all other replicas, and after other cpnd apply these change, they will send CPA_EVT_ND2A_CKPT_ARRIVAL_NTFY to their local client for data change, but does not pass in change data, just section ID. Do you think it is possible to pass the data along with the section to agents, it will be more efficient, because agent does not need additional read? With patch you mentioned below, if read_size not passed as zero, will the agent get the changed data? Thanks. Ted From: A V Mahesh [mailto:[email protected]] Sent: Tuesday, May 26, 2015 11:35 AM To: Yao Cheng LIANG; [email protected]<mailto:[email protected]> Subject: Re: [users] ckptnd memory leak(very serious) Hi Ted, It was committed on `Fri May 15` please take the patch from community . changeset: 6556:41a382748df9 user: A V Mahesh <[email protected]><mailto:[email protected]> date: Fri May 15 13:11:31 2015 +0530 summary: cpsv : fix CPA & CPND memleak [#1373] -AVM On 5/25/2015 6:00 PM, Yao Cheng LIANG wrote: Dear AVM, Do you have a plan for this patch? Ted From: A V Mahesh [mailto:[email protected]] Sent: 2015年5月14日 17:31 To: Xiao Dong WANG; Ou Jia Dong; Jesse Y. AI; Yao Cheng LIANG; [email protected]<mailto:[email protected]> Cc: Mui Ka Ho Alex Subject: Re: [users] ckptnd memory leak(very serious) Hi Xiaodong, Your are right It looks complex to fix , currently I am working on patch , as a work around comment out below as your already done and proceed your testing . //send_evt.info.cpa.info.arr_msg.ckpt_data->data = NULL; //send_evt.info.cpa.info.arr_msg.ckpt_data->readSize = 0; -AVM ------------------------------------------------------------------------------ One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y _______________________________________________ Opensaf-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/opensaf-users
