Better "final read from parent" algorithm in rewriteAppendOnlyFile*(.

We now wait up to 1 second for diff data to come from the parent,
however we use poll(2) to wait for more data, and use a counter of
contiguous failures to get data for N times (set to 20 experimentally
after different tests) as an early stop condition to avoid wasting 1
second when the write traffic is too low.
This commit is contained in:
antirez 2014-07-05 15:40:08 +02:00
parent 895409ca78
commit 2de5bab368

View File

@ -1059,17 +1059,20 @@ int rewriteAppendOnlyFile(char *filename) {
/* Read again a few times to get more data from the parent.
* We can't read forever (the server may receive data from clients
* fater than it is able to send data to the child), so we try to read
* some more data in a loop and wait a bit if data is not available, but
* don't wait more than 100 ms total. */
* some more data in a loop as soon as there is a good chance more data
* will come. If it looks like we are wasting time, we abort (this
* happens after 20 ms without new data). */
int nodata = 0;
for (j = 0; j < 100; j++) {
ssize_t nread_from_parent = aofReadDiffFromParent();
printf("%lld\n", (long long) nread_from_parent);
if (nread_from_parent == 0) {
mstime_t start = mstime();
while(mstime()-start < 1000 && nodata < 20) {
if (aeWait(server.aof_pipe_read_data_from_parent, AE_READABLE, 1) <= 0)
{
nodata++;
if (nodata == 10) break;
usleep(10000);
continue;
}
nodata = 0; /* Start counting from zero, we stop on N *contiguous*
timeouts. */
aofReadDiffFromParent();
}
/* Ask the master to stop sending diffs. */