http: cork/uncork before flushing pipelined res

[indutny]

Make sure that the pipelined response data will be written as less TCP
packets as possible.

Benchmark results (3 runs):

http/chunked.js num=1 size=1 c=100: ./node-after: 2358.9 ./node-before: 2374.1 ................................... -0.64%                                                                           [87/709]
http/chunked.js num=1 size=64 c=100: ./node-after: 2372.7 ./node-before: 2373.2 .................................. -0.02%
http/chunked.js num=1 size=256 c=100: ./node-after: 2374.2 ./node-before: 2367.9 .................................. 0.27%
http/chunked.js num=4 size=1 c=100: ./node-after: 2363.1 ./node-before: 2356 ...................................... 0.30%
http/chunked.js num=4 size=64 c=100: ./node-after: 2370.7 ./node-before: 2358.6 ................................... 0.51%
http/chunked.js num=4 size=256 c=100: ./node-after: 2364.8 ./node-before: 2355.7 .................................. 0.39%
http/chunked.js num=8 size=1 c=100: ./node-after: 2231.5 ./node-before: 2332.9 ................................... -4.34%
http/chunked.js num=8 size=64 c=100: ./node-after: 2253.1 ./node-before: 2357.3 .................................. -4.42%
http/chunked.js num=8 size=256 c=100: ./node-after: 2346.7 ./node-before: 2300.2 .................................. 2.02%
http/chunked.js num=16 size=1 c=100: ./node-after: 2248.2 ./node-before: 2076.3 ................................... 8.28%
http/chunked.js num=16 size=64 c=100: ./node-after: 2223.2 ./node-before: 2052.4 .................................. 8.32%
http/chunked.js num=16 size=256 c=100: ./node-after: 1928.6 ./node-before: 2133 .................................. -9.58%
http/client-request-body.js dur=5 type=asc bytes=32 method=write: ./node-after: 3296.9 ./node-before: 49.987 ... 6495.58%
http/client-request-body.js dur=5 type=asc bytes=32 method=end: ./node-after: 3308.6 ./node-before: 49.598 ..... 6570.96%
http/client-request-body.js dur=5 type=asc bytes=256 method=write: ./node-after: 3462.5 ./node-before: 49.595 .. 6881.48%
http/client-request-body.js dur=5 type=asc bytes=256 method=end: ./node-after: 3520 ./node-before: 49.599 ...... 6997.02%
http/client-request-body.js dur=5 type=asc bytes=1024 method=write: ./node-after: 3384.3 ./node-before: 49.59 .. 6724.59%
http/client-request-body.js dur=5 type=asc bytes=1024 method=end: ./node-after: 3409.9 ./node-before: 49.594 ... 6775.71%
http/client-request-body.js dur=5 type=utf bytes=32 method=write: ./node-after: 3373.4 ./node-before: 49.597 ... 6701.50%
http/client-request-body.js dur=5 type=utf bytes=32 method=end: ./node-after: 3319.9 ./node-before: 49.991 ..... 6540.99%
http/client-request-body.js dur=5 type=utf bytes=256 method=write: ./node-after: 3606.5 ./node-before: 49.994 .. 7113.99%
http/client-request-body.js dur=5 type=utf bytes=256 method=end: ./node-after: 3211 ./node-before: 49.592 ...... 6374.84%
http/client-request-body.js dur=5 type=utf bytes=1024 method=write: ./node-after: 2963.8 ./node-before: 49.585 . 5877.28%
http/client-request-body.js dur=5 type=utf bytes=1024 method=end: ./node-after: 3298.2 ./node-before: 49.986 ... 6498.27%
http/client-request-body.js dur=5 type=buf bytes=32 method=write: ./node-after: 3454.9 ./node-before: 49.99 .... 6811.27%
http/client-request-body.js dur=5 type=buf bytes=32 method=end: ./node-after: 3512.3 ./node-before: 49.599 ..... 6981.38%
http/client-request-body.js dur=5 type=buf bytes=256 method=write: ./node-after: 3459.6 ./node-before: 49.595 .. 6875.69%
http/client-request-body.js dur=5 type=buf bytes=256 method=end: ./node-after: 3642.6 ./node-before: 49.99 ..... 7186.51%
http/client-request-body.js dur=5 type=buf bytes=1024 method=write: ./node-after: 3522.2 ./node-before: 49.989 . 6945.95%
http/client-request-body.js dur=5 type=buf bytes=1024 method=end: ./node-after: 3551 ./node-before: 49.597 ..... 7059.66%
http/cluster.js type=bytes length=4 c=50: ./node-after: 17203 ./node-before: 14735 ............................... 16.75%
http/cluster.js type=bytes length=4 c=500: ./node-after: 15383 ./node-before: 15141 ............................... 1.60%
http/cluster.js type=bytes length=1024 c=50: ./node-after: 15554 ./node-before: 15133 ............................. 2.78%
http/cluster.js type=bytes length=1024 c=500: ./node-after: 14598 ./node-before: 14316 ............................ 1.97%
http/cluster.js type=bytes length=102400 c=50: ./node-after: 844.36 ./node-before: 887.37 ........................ -4.85%
http/cluster.js type=bytes length=102400 c=500: ./node-after: 760.55 ./node-before: 850.88 ...................... -10.62%
http/cluster.js type=buffer length=4 c=50: ./node-after: 14737 ./node-before: 15454 .............................. -4.64%
http/cluster.js type=buffer length=4 c=500: ./node-after: 13455 ./node-before: 13824 ............................. -2.67%
http/cluster.js type=buffer length=1024 c=50: ./node-after: 14033 ./node-before: 14390 ........................... -2.48%
http/cluster.js type=buffer length=1024 c=500: ./node-after: 14124 ./node-before: 14843 .......................... -4.85%
http/cluster.js type=buffer length=102400 c=50: ./node-after: 12275 ./node-before: 11311 .......................... 8.52%
http/cluster.js type=buffer length=102400 c=500: ./node-after: 11404 ./node-before: 12162 ........................ -6.23%
http/end-vs-write-end.js type=asc kb=64 c=100 method=write: ./node-after: 4567.7 ./node-before: 4516.2 ............ 1.14%
http/end-vs-write-end.js type=asc kb=64 c=100 method=end: ./node-after: 4582.2 ./node-before: 4417.7 .............. 3.72%
http/end-vs-write-end.js type=asc kb=128 c=100 method=write: ./node-after: 2688.4 ./node-before: 2932.7 .......... -8.33%
http/end-vs-write-end.js type=asc kb=128 c=100 method=end: ./node-after: 2985.8 ./node-before: 3066.6 ............ -2.63%
http/end-vs-write-end.js type=asc kb=256 c=100 method=write: ./node-after: 1733.4 ./node-before: 1703.8 ........... 1.74%
http/end-vs-write-end.js type=asc kb=256 c=100 method=end: ./node-after: 1805.5 ./node-before: 1656.4 ............. 9.00%
http/end-vs-write-end.js type=asc kb=1024 c=100 method=write: ./node-after: 361.9 ./node-before: 362.04 .......... -0.04%
http/end-vs-write-end.js type=asc kb=1024 c=100 method=end: ./node-after: 362.07 ./node-before: 357.1 ............. 1.39%
http/end-vs-write-end.js type=utf kb=64 c=100 method=write: ./node-after: 3483.3 ./node-before: 3697.7 ........... -5.80%
http/end-vs-write-end.js type=utf kb=64 c=100 method=end: ./node-after: 3986.4 ./node-before: 3818 ................ 4.41%
http/end-vs-write-end.js type=utf kb=128 c=100 method=write: ./node-after: 2320.3 ./node-before: 2177.1 ........... 6.58%
http/end-vs-write-end.js type=utf kb=128 c=100 method=end: ./node-after: 2532.2 ./node-before: 2437.6 ............. 3.88%
http/end-vs-write-end.js type=utf kb=256 c=100 method=write: ./node-after: 1388 ./node-before: 1411.2 ............ -1.65%
http/end-vs-write-end.js type=utf kb=256 c=100 method=end: ./node-after: 1622.8 ./node-before: 1367.6 ............ 18.67%
http/end-vs-write-end.js type=utf kb=1024 c=100 method=write: ./node-after: 412.43 ./node-before: 415.73 ......... -0.79%
http/end-vs-write-end.js type=utf kb=1024 c=100 method=end: ./node-after: 385.78 ./node-before: 399.51 ........... -3.44%
http/end-vs-write-end.js type=buf kb=64 c=100 method=write: ./node-after: 7876.5 ./node-before: 6939.5 ........... 13.50%
http/end-vs-write-end.js type=buf kb=64 c=100 method=end: ./node-after: 7916.3 ./node-before: 7418.2 .............. 6.71%
http/end-vs-write-end.js type=buf kb=128 c=100 method=write: ./node-after: 6000.9 ./node-before: 6474.2 .......... -7.31%
http/end-vs-write-end.js type=buf kb=128 c=100 method=end: ./node-after: 6715.9 ./node-before: 6786.8 ............ -1.04%
http/end-vs-write-end.js type=buf kb=256 c=100 method=write: ./node-after: 5026.9 ./node-before: 5400.1 .......... -6.91%
http/end-vs-write-end.js type=buf kb=256 c=100 method=end: ./node-after: 5380.5 ./node-before: 5861.6 ............ -8.21%
http/end-vs-write-end.js type=buf kb=1024 c=100 method=write: ./node-after: 2646.3 ./node-before: 2466.3 .......... 7.30%
http/end-vs-write-end.js type=buf kb=1024 c=100 method=end: ./node-after: 2518.6 ./node-before: 2407.1 ............ 4.63%
http/_chunky_http_client.js len=1 num=5 type=send: ./node-after: 64.799 ./node-before: 64.931 .................... -0.20%
http/_chunky_http_client.js len=1 num=50 type=send: ./node-after: 173.88 ./node-before: 187.61 ................... -7.32%
http/_chunky_http_client.js len=1 num=500 type=send: ./node-after: 243.19 ./node-before: 233.04 ................... 4.36%
http/_chunky_http_client.js len=1 num=2000 type=send: ./node-after: 239.9 ./node-before: 247.48 .................. -3.06%
http/_chunky_http_client.js len=4 num=5 type=send: ./node-after: 135.87 ./node-before: 161.15 ................... -15.69%
http/_chunky_http_client.js len=4 num=50 type=send: ./node-after: 176.15 ./node-before: 131.64 ................... 33.81%
http/_chunky_http_client.js len=4 num=500 type=send: ./node-after: 219.26 ./node-before: 237.87 .................. -7.82%
http/_chunky_http_client.js len=4 num=2000 type=send: ./node-after: 271.31 ./node-before: 224.63 ................. 20.78%
http/_chunky_http_client.js len=8 num=5 type=send: ./node-after: 120.67 ./node-before: 138.4 .................... -12.81%
http/_chunky_http_client.js len=8 num=50 type=send: ./node-after: 187.72 ./node-before: 202.46 ................... -7.28%
http/_chunky_http_client.js len=8 num=500 type=send: ./node-after: 222.44 ./node-before: 219.25 ................... 1.45%
http/_chunky_http_client.js len=8 num=2000 type=send: ./node-after: 275.45 ./node-before: 208.39 ................. 32.18%
http/_chunky_http_client.js len=16 num=5 type=send: ./node-after: 128.75 ./node-before: 114.76 ................... 12.19%
http/_chunky_http_client.js len=16 num=50 type=send: ./node-after: 169.7 ./node-before: 199.12 .................. -14.77%
http/_chunky_http_client.js len=16 num=500 type=send: ./node-after: 203.63 ./node-before: 208.51 ................. -2.34%
http/_chunky_http_client.js len=16 num=2000 type=send: ./node-after: 244.63 ./node-before: 235.49 ................. 3.88%
http/_chunky_http_client.js len=32 num=5 type=send: ./node-after: 104.66 ./node-before: 48.936 .................. 113.87%
http/_chunky_http_client.js len=32 num=50 type=send: ./node-after: 202.27 ./node-before: 162.8 ................... 24.25%
http/_chunky_http_client.js len=32 num=500 type=send: ./node-after: 242.88 ./node-before: 241.29 .................. 0.66%
http/_chunky_http_client.js len=32 num=2000 type=send: ./node-after: 239.01 ./node-before: 206.21 ................ 15.91%
http/_chunky_http_client.js len=64 num=5 type=send: ./node-after: 143.12 ./node-before: 48.008 .................. 198.12%
http/_chunky_http_client.js len=64 num=50 type=send: ./node-after: 222.88 ./node-before: 135.39 .................. 64.62%
http/_chunky_http_client.js len=64 num=500 type=send: ./node-after: 227.41 ./node-before: 240.4 .................. -5.40%
http/_chunky_http_client.js len=64 num=2000 type=send: ./node-after: 264.16 ./node-before: 251.04 ................. 5.23%
http/_chunky_http_client.js len=128 num=5 type=send: ./node-after: 46.779 ./node-before: 44.096 ................... 6.09%
http/_chunky_http_client.js len=128 num=50 type=send: ./node-after: 61.551 ./node-before: 188.66 ................ -67.37%
http/_chunky_http_client.js len=128 num=500 type=send: ./node-after: 217.13 ./node-before: 266.04 ............... -18.38%
http/_chunky_http_client.js len=128 num=2000 type=send: ./node-after: 232.96 ./node-before: 259.63 .............. -10.28%
http/simple.js type=bytes length=4 chunks=0 c=50: ./node-after: 9245.3 ./node-before: 9118.4 ...................... 1.39%
http/simple.js type=bytes length=4 chunks=0 c=500: ./node-after: 9255.5 ./node-before: 9152.4 ..................... 1.13%
http/simple.js type=bytes length=4 chunks=1 c=50: ./node-after: 8326.3 ./node-before: 8520.3 ..................... -2.28%
http/simple.js type=bytes length=4 chunks=1 c=500: ./node-after: 7902.2 ./node-before: 7738.2 ..................... 2.12%
http/simple.js type=bytes length=4 chunks=4 c=50: ./node-after: 1185.1 ./node-before: 1189.9 ..................... -0.40%
http/simple.js type=bytes length=4 chunks=4 c=500: ./node-after: 5658.5 ./node-before: 5339.8 ..................... 5.97%
http/simple.js type=bytes length=1024 chunks=0 c=50: ./node-after: 8493 ./node-before: 8849.1 .................... -4.03%
http/simple.js type=bytes length=1024 chunks=0 c=500: ./node-after: 8121.5 ./node-before: 8035 .................... 1.08%
http/simple.js type=bytes length=1024 chunks=1 c=50: ./node-after: 6813.1 ./node-before: 6005.2 .................. 13.45%
http/simple.js type=bytes length=1024 chunks=1 c=500: ./node-after: 6371.4 ./node-before: 7003.9 ................. -9.03%
http/simple.js type=bytes length=1024 chunks=4 c=50: ./node-after: 1189.9 ./node-before: 1191.7 .................. -0.15%
http/simple.js type=bytes length=1024 chunks=4 c=500: ./node-after: 4826.9 ./node-before: 5148.8 ................. -6.25%
http/simple.js type=bytes length=102400 chunks=0 c=50: ./node-after: 397.49 ./node-before: 435.99 ................ -8.83%
http/simple.js type=bytes length=102400 chunks=0 c=500: ./node-after: 410.8 ./node-before: 445.48 ................ -7.78%
http/simple.js type=bytes length=102400 chunks=1 c=50: ./node-after: 235.8 ./node-before: 251.15 ................. -6.11%
http/simple.js type=bytes length=102400 chunks=1 c=500: ./node-after: 254.27 ./node-before: 253.64 ................ 0.25%
http/simple.js type=bytes length=102400 chunks=4 c=50: ./node-after: 1953.3 ./node-before: 2025.7 ................ -3.57%
http/simple.js type=bytes length=102400 chunks=4 c=500: ./node-after: 1822.2 ./node-before: 1971.7 ............... -7.58%
http/simple.js type=buffer length=4 chunks=0 c=50: ./node-after: 7697.4 ./node-before: 8308.6 .................... -7.36%
http/simple.js type=buffer length=4 chunks=0 c=500: ./node-after: 7722 ./node-before: 7645.4 ...................... 1.00%
http/simple.js type=buffer length=4 chunks=1 c=50: ./node-after: 7423 ./node-before: 7111.4 ....................... 4.38%
http/simple.js type=buffer length=4 chunks=1 c=500: ./node-after: 7565.9 ./node-before: 7081.1 .................... 6.85%
http/simple.js type=buffer length=4 chunks=4 c=50: ./node-after: 7343 ./node-before: 6543.7 ...................... 12.21%
http/simple.js type=buffer length=4 chunks=4 c=500: ./node-after: 6732.3 ./node-before: 6857.2 ................... -1.82%
http/simple.js type=buffer length=1024 chunks=0 c=50: ./node-after: 8176.5 ./node-before: 8441.2 ................. -3.14%
http/simple.js type=buffer length=1024 chunks=0 c=500: ./node-after: 7648.5 ./node-before: 8213.4 ................ -6.88%
http/simple.js type=buffer length=1024 chunks=1 c=50: ./node-after: 8139.4 ./node-before: 7969 .................... 2.14%
http/simple.js type=buffer length=1024 chunks=1 c=500: ./node-after: 7342.6 ./node-before: 7223.7 ................. 1.65%
http/simple.js type=buffer length=1024 chunks=4 c=50: ./node-after: 6663.5 ./node-before: 7265.7 ................. -8.29%
http/simple.js type=buffer length=1024 chunks=4 c=500: ./node-after: 5862.8 ./node-before: 6467.6 ................ -9.35%
http/simple.js type=buffer length=102400 chunks=0 c=50: ./node-after: 6431.9 ./node-before: 6522.8 ............... -1.39%
http/simple.js type=buffer length=102400 chunks=0 c=500: ./node-after: 6000.1 ./node-before: 6609.7 .............. -9.22%
http/simple.js type=buffer length=102400 chunks=1 c=50: ./node-after: 6263.9 ./node-before: 6122.4 ................ 2.31%
http/simple.js type=buffer length=102400 chunks=1 c=500: ./node-after: 6340.6 ./node-before: 5743 ................ 10.40%
http/simple.js type=buffer length=102400 chunks=4 c=50: ./node-after: 5644.4 ./node-before: 5665.6 ............... -0.37%
http/simple.js type=buffer length=102400 chunks=4 c=500: ./node-after: 5786.2 ./node-before: 5589.8 ............... 3.51%

cc @trevnorris (we need performance WG)

[mscdex]

LGTM

[indutny]

May I ask anybody to re-run the benchmarks to confirm that these -60% a just spurious?

[trevnorris]

@indutny The http benchmarks vary a lot. That difference appears to just be some test variance.

OutgoingMessage#write() is already calling cork() in the case where data is a buffer. I remember trying to generalize it to all cases but for some reason had an issue with that. My worry is that you'll be uncorking on a connection that's still corked if a user writes both buffers and strings. Think it could be generalized to all cases?

The other difference is that write() calls uncork() on the nextTick() instead of when the data is done being written. (this may only affect the case of _writeRaw()).

[indutny]

@trevnorris cork calls are balanced, no spurious uncorking will happen...

[trevnorris]

Cool. Then LGTM.

[indutny]

CI: https://ci.nodejs.org/job/node-test-pull-request/436/

[trevnorris]

Tests look good. Only external string bytes is failing.

[indutny]

CI is green

[indutny]

Landed in 184f09d, thank you everyone!