Speed up BUFFER increases (minimize reallocs)

[ktsaou]

Optimize BUFFER increases for large web requests.

This PR makes buffer increases about 25% faster for large buffers.

In the following test case, using the BUFFER of the master, it finishes in 65 ms. With this PR it finishes in 50 ms.

Test:

int main(int argc, char **argv) {
    BUFFER *wb = buffer_create(1024);

    const char *string = "012345678901234567890123456789012345678901234567890123456789";
    size_t size_of_string = strlen(string);
    size_t total = 0;

    usec_t start = now_realtime_usec();
    while(total < 40*1024*1024) {
        buffer_sprintf(wb, "%s", string);
        total += size_of_string;
    }
    usec_t end = now_realtime_usec();
    printf("Buffer of %zu bytes finished in %llu usec\n", buffer_strlen(wb), end-start);

    log_allocations();

    buffer_free(wb);
    exit(0);
}

Test results from master:

Buffer of 41943060 bytes finished in 65844 usec
MAIN MEMORY ALLOCATIONS: (0697@daemon/main.c:main): Allocated 40961 KiB (+41944078 B), mmapped 0 KiB (+0 B): : malloc 1 (+1), calloc 1 (+1), realloc 40920 (+40920), strdup 0 (+0), free 0 (+0)

Test results from this PR:

Buffer of 41943060 bytes finished in 50036 usec
MAIN MEMORY ALLOCATIONS: (0697@daemon/main.c:main): Allocated 49152 KiB (+50331702 B), mmapped 0 KiB (+0 B): : malloc 1 (+1), calloc 1 (+1), realloc 15 (+15), strdup 0 (+0), free 0 (+0)

---

real-life effect

Running a parallel test with on a context with 1000 charts, 100 dimensions each, like this:

siege --verbose --concurrent=255 --benchmark --reps=1 "http://localhost:19999/api/v1/data?context=example.random&before=0&after=-60&points=60&options=jsonwrap"

This PR provides a 10% benefit.

The master:

Transactions:                255 hits
Availability:             100.00 %
Elapsed time:             131.46 secs
Data transferred:        2384.86 MB
Response time:            110.83 secs
Transaction rate:           1.94 trans/sec
Throughput:            18.14 MB/sec
Concurrency:              214.98
Successful transactions:         255
Failed transactions:               0
Longest transaction:          131.46
Shortest transaction:          93.43

This PR:

Transactions:                255 hits
Availability:             100.00 %
Elapsed time:             119.09 secs
Data transferred:        2384.76 MB
Response time:             97.45 secs
Transaction rate:           2.14 trans/sec
Throughput:            20.02 MB/sec
Concurrency:              208.66
Successful transactions:         255
Failed transactions:               0
Longest transaction:          119.09
Shortest transaction:          83.44

[thiagoftsm]

I tested this PR and I compare with master branch, and I can confirm improvement in our stats.
I also observe that cloud was working fine and the host was visible all the time.
LGTM!