| James, A lock can be obtained in the parse, plan and execute step, depending on cache, state and type of object. A LWLock is a spinlock, a low level access mechanism that is supposed to be extremely quickly. It is used to serialise access to elementary structures mostly for changes. A Lock is an higher level lock that is much more sophisticated, contains multiple states and can order multiple requests. It is used to safeguard transaction intention for objects. The wait LWLock:LockManager is documented to have two common reasons: too many locks being acquired, exceeding the fastpath slots number (16) and/or exceeding CPU capacity. What is happening if you are waiting for a LWLock is that the number of processes trying to access the structure (the lock manager) is higher than one. Because the LWLock is meant to be held so briefly that there should no waiting, it means that if you are waiting for it, there must be a reason it’s held so long. An obvious reason for holding a LWLock too long is if there are more tasks on the OS than CPU’s, as Laurenz indicates. If such a situation happens, it’s possible a tasks is put off CPU by the operating system whilst holding the LWLock, which will greatly increase the time waiting for it, because the LWLock can only be released if the task manages to get back on CPU. Regards, Frits Hoogland
|
