This is a small note describing how Oracle implemented the situation which is covered by the db file parallel read wait event. This events happens if Oracle knows it must read multiple blocks which are not adjacent (thus from different random files and locations), and cannot continue processing with the result of a single block. In other words: if it cannot process something after reading a single block (otherwise Oracle will read a single block visible by the wait ‘db file sequential read’).
This is how it shows up if you enable sql trace:
WAIT #139658359011504: nam='db file parallel read' ela= 69997140 files=1 blocks=70 requests=70 obj#=76227 tim=1373200106669612
What this shows, is Oracle issuing a request for 70 blocks. This has an interesting implication for monitoring and looking at the time spend on the event ‘db file parallel read': if you don’t know the number of blocks for which an IO request is issued, it’s impossible to say something about the time. So just monitoring or looking a cumulative time spend in ‘db file parallel read’ doesn’t say anything about IO latency, it only tells something about where the Oracle process did spend its time on.
How did Oracle implement this? This is obviously port specific (which means the implementation will be different on different operating systems). My test environment is Oracle Linux 6u3 X64, Oracle 126.96.36.199 64 bit using ASM.
This is how the requests are asynchronously submitted to the operating system:
Breakpoint 2, io_submit (ctx=0x7f04c0c8d000, nr=70, iocbs=0x7fff86d965f0) at io_submit.c:23
So all the IO requests are submitted in one go!
After the IO requests are submitted (which is not covered by a wait, which makes sense, because the io_submit call is/is supposed to be non blocking.
Next Oracle waits for ALL the IOs to finish, covered by the ‘db file parallel read’ wait event:
Breakpoint 13, 0x0000000008f9a652 in kslwtbctx () Breakpoint 1, io_getevents_0_4 (ctx=0x7f04c0c8d000, min_nr=70, nr=128, events=0x7fff86d9c798, timeout=0x7fff86d9d7a0) at io_getevents.c:46 Breakpoint 14, 0x0000000008fa1334 in kslwtectx ()
We see kslwtbctx which indicates the start of a waitevent, then a io_getevents call:
‘ctx’ is the IO context. This is how Linux keeps track of groups of asynchronous IO requests.
‘min_nr’ is the minimal number of requests that must be ready for this call to succeed, this call will wait until ‘timeout’ is reached. io_getevents will just peek if ‘timeout’ is set to zero.
‘nr’ is the maximal number of requests that io_getevents will “fetch”.
‘events’ is a struct (table like structure) that holds the information about the iocb’s (IO control blocks) of the requests.
‘timeout’ is a struct that sets the timeout of this call. For Oracle IO I see timeout being 600 (seconds) most of the time.
The last line show kslwtectx indicating that the wait has ended.