>So, it seems like you're basically using malloc to work around the
>fact that a palloc failure is an error, and we can't throw an error in
>a critical section. I don't think that's good; we want all of our
>allocations, as far as possible, to be tracked via palloc. It might
>be a good idea to add a new variant of palloc or MemoryContextAlloc
>that returns NULL on failure instead of throwing an error; I've wanted
>that once or twice. But in this particular case, I'm not quite seeing
>why it should be necessary
I am using malloc to return NULL in case of failure and proceed without compression of FPW ,if it returns NULL.
Proceeding without compression seems to be more accurate than throwing an error and exiting because of failure to allocate memory for compression.
>the number of backup blocks per record is
>limited to some pretty small number, so it ought to be possible to
>preallocate enough memory to compress them all, perhaps just by
>declaring a global variable like char wal_compression_space[8192]; or
>whatever.
In the updated patch a static global variable is added to which memory is allocated from heap using malloc outside critical section. The size of the memory block is 4 * BkpBlock header + 4 * BLCKSZ.