digitalmars.D.learn - Comparison : mysql-native + asdf and hunt-database + asdf
- Vino (22/22) Nov 05 2020 Hi All,
- frame (3/9) Nov 06 2020 Interesting but I think you should provide the test code for
- Andre Pany (9/31) Nov 07 2020 While doing performance measurements you should do each test
Hi All, We recently tested the below components and the test results are as below, even though hunt-database is faster than mysql-native it is hard to use this package as it lacks on documentation, non of the example provided in the documentation works, one has to go through the code and find how things works to use this package. We have raised this issue in the github and it been 3 day's still no update and the last documentation (README) was updated 14 month ago, but the latest release of this package was released on 2020-Oct-11, version 2.1.9. May I know whether is there is any plan for updating the documentation if so may know when would the documentation be updated, so that we can explore other features of hunt packages (hunt-framework vs vib.d framework). Component : mysql-native + asdf Executable size : 17 MB Execution time : 10 secs, 189 ms, 919 μs, and 3 hnsecs Component : hunt-database + asdf Executable size : 81 MB Execution time : 5 secs, 916 ms, 418 μs, and 3 hnsecs From, Vino.B
Nov 05 2020
frame <frame86 live.com> On Friday, 6 November 2020 at 04:58:05 UTC, Vino wrote:Component : mysql-native + asdf Executable size : 17 MB Execution time : 10 secs, 189 ms, 919 μs, and 3 hnsecs Component : hunt-database + asdf Executable size : 81 MB Execution time : 5 secs, 916 ms, 418 μs, and 3 hnsecsInteresting but I think you should provide the test code for representative results?
Nov 06 2020
On Friday, 6 November 2020 at 04:58:05 UTC, Vino wrote:Hi All, We recently tested the below components and the test results are as below, even though hunt-database is faster than mysql-native it is hard to use this package as it lacks on documentation, non of the example provided in the documentation works, one has to go through the code and find how things works to use this package. We have raised this issue in the github and it been 3 day's still no update and the last documentation (README) was updated 14 month ago, but the latest release of this package was released on 2020-Oct-11, version 2.1.9. May I know whether is there is any plan for updating the documentation if so may know when would the documentation be updated, so that we can explore other features of hunt packages (hunt-framework vs vib.d framework). Component : mysql-native + asdf Executable size : 17 MB Execution time : 10 secs, 189 ms, 919 μs, and 3 hnsecs Component : hunt-database + asdf Executable size : 81 MB Execution time : 5 secs, 916 ms, 418 μs, and 3 hnsecs From, Vino.BWhile doing performance measurements you should do each test multiple time (at least 5 times). There could be for example an effect that executing a db query the first time is a lot slower than afterwards. Maybe mysql native is slower, but maybe this isn't the case, just the way performance measurements was done was incorrectly. Kind regards Andre
Nov 07 2020
On Saturday, 7 November 2020 at 12:29:46 UTC, Andre Pany wrote:On Friday, 6 November 2020 at 04:58:05 UTC, Vino wrote:Hi Andre, The benchmark was performed with 100 executions as below void main() { void f() { Array!string[string] data = getdata("TEST"); foreach(i; data) { writeln(i[]); } } auto r = benchmark!(f)(100); Duration t = r[0]; writeln(t); }[...]While doing performance measurements you should do each test multiple time (at least 5 times). There could be for example an effect that executing a db query the first time is a lot slower than afterwards. Maybe mysql native is slower, but maybe this isn't the case, just the way performance measurements was done was incorrectly. Kind regards Andre
Nov 07 2020
On Saturday, 7 November 2020 at 14:16:46 UTC, Vino wrote:On Saturday, 7 November 2020 at 12:29:46 UTC, Andre Pany wrote:Hi All, Upon further testing, below are the observations, based on different compliers (DMD/LDC), using LDC the size of the executable is reduced but the run time is increased from 5 to 6 sec , the document states that if we use the complier option dub --"build=release --compiler=ldmd2" it would enhance the performance but as per our analysis we do not see and improvement, below is the runtime LDC : 6 secs, 136 ms, 97 μs, and 3 hnsecs LDC(dub --build=release --compiler=ldmd2): 6 secs, 203 ms, 283 μs, and 6 hnsecs Observations ******************** Component : mysql-native + asdf Complier : DMD Size : 17MB Duration : 10 secs, 189 ms, 919 μs, and 3 hnsecs Component : mysql-native + asdf Complier : LDC Size : 20MB Duration : 10 secs, 526 ms, 350 μs, and 6 hnsecs Component : mysql-native + asdf Complier : LDC(dub --build=release --compiler=ldmd2) Size : 3.7MB Duration : 10 secs, 411 ms, 793 μs, and 8 hnsecs ************************************************************* Component : hunt-database + asdf Complier : DMD Size : 81MB Duration : 5 secs, 916 ms, 418 μs, and 3 hnsecs Component : hunt-database + asdf Complier : LDC Size : 50MB Duration : 6 secs, 136 ms, 97 μs, and 3 hnsecs Component : hunt-database + asdf Complier : LDC(dub --build=release --compiler=ldmd2) Size : 17MB Duration : 6 secs, 136 ms, 97 μs, and 3 hnsecs From, Vino.BOn Friday, 6 November 2020 at 04:58:05 UTC, Vino wrote:Hi Andre, The benchmark was performed with 100 executions as below void main() { void f() { Array!string[string] data = getdata("TEST"); foreach(i; data) { writeln(i[]); } } auto r = benchmark!(f)(100); Duration t = r[0]; writeln(t); }[...]While doing performance measurements you should do each test multiple time (at least 5 times). There could be for example an effect that executing a db query the first time is a lot slower than afterwards. Maybe mysql native is slower, but maybe this isn't the case, just the way performance measurements was done was incorrectly. Kind regards Andre
Nov 07 2020
On Saturday, 7 November 2020 at 14:38:35 UTC, Vino wrote:On Saturday, 7 November 2020 at 14:16:46 UTC, Vino wrote:Hi All, After further analysis we suspect that the issue is at the package std.net.curl the flow of the program is as below hunt.database : queries a single table(12 column's and 3 rows) which contains the 3 api url details(getApiconf) std.net.curl : call those api's (parallel) and the send the data to json parser asdf : Parse the output of the api's and store the data in a array Array!string[string] data Code of the API calls import common.GetApiconf; import core.stdc.stdlib: exit; import hunt.database; import std.algorithm: joiner, sort; import std.container.array; import std.conv: to; import std.net.curl : get, HTTP, CurlOption; import std.parallelism: parallel; import std.typecons: Tuple, tuple; void main () { Array!(Tuple!(int,string)) apidata; Row[] result; result = getApiconf("TST"); foreach(i, k; parallel(result,1)) { string apihost = result[i][0].get!(string); int apiport = result[i][1].get!(int); string apiuser = result[i][2].get!(string); string apipass = result[i][3].get!(string); int seq = result[i][4].get!int; string apiuri = result[i][5].get!(string); int connecttimeout = result[i][6].get!(int); int tcp_nodelay = result[i][7].get!(int); int http_version = result[i][8].get!(int); int sslversion = result[i][9].get!(int); int use_ssl = result[i][10].get!(int); int ssl_verifypeer = result[i][11].get!(int); string url = to!string(([apihost, ":", to!string(apiport), apiuri]).joiner); string usrpass = to!string(([apiuser, ":", apipass]).joiner); auto https = HTTP(); https.handle.set(CurlOption.userpwd, usrpass); https.handle.set(CurlOption.connecttimeout, connecttimeout); https.handle.set(CurlOption.tcp_nodelay, tcp_nodelay); https.handle.set(CurlOption.http_version, http_version); https.handle.set(CurlOption.sslversion, sslversion); https.handle.set(CurlOption.use_ssl, use_ssl); https.handle.set(CurlOption.ssl_verifypeer, ssl_verifypeer); https.handle.set(CurlOption.url, url); https.method(HTTP.Method.get); https.StatusLine st; https.onReceiveStatusLine = (https.StatusLine st) { if (st.code != 200) { throw new Exception(st.reason); } }; ubyte[] content; https.onReceive = (ubyte[] data) { content ~= data; return data.length; }; https.perform(); scope(failure) { https.shutdown; exit(-4); } scope(exit) https.shutdown; apidata.insert(tuple(seq, cast(string) content)); } writeln(apidata[].sort); } From, Vino.BOn Saturday, 7 November 2020 at 12:29:46 UTC, Andre Pany wrote:Hi All, Upon further testing, below are the observations, based on different compliers (DMD/LDC), using LDC the size of the executable is reduced but the run time is increased from 5 to 6 sec , the document states that if we use the complier option dub --"build=release --compiler=ldmd2" it would enhance the performance but as per our analysis we do not see and improvement, below is the runtime LDC : 6 secs, 136 ms, 97 μs, and 3 hnsecs LDC(dub --build=release --compiler=ldmd2): 6 secs, 203 ms, 283 μs, and 6 hnsecs Observations ******************** Component : mysql-native + asdf Complier : DMD Size : 17MB Duration : 10 secs, 189 ms, 919 μs, and 3 hnsecs Component : mysql-native + asdf Complier : LDC Size : 20MB Duration : 10 secs, 526 ms, 350 μs, and 6 hnsecs Component : mysql-native + asdf Complier : LDC(dub --build=release --compiler=ldmd2) Size : 3.7MB Duration : 10 secs, 411 ms, 793 μs, and 8 hnsecs ************************************************************* Component : hunt-database + asdf Complier : DMD Size : 81MB Duration : 5 secs, 916 ms, 418 μs, and 3 hnsecs Component : hunt-database + asdf Complier : LDC Size : 50MB Duration : 6 secs, 136 ms, 97 μs, and 3 hnsecs Component : hunt-database + asdf Complier : LDC(dub --build=release --compiler=ldmd2) Size : 17MB Duration : 6 secs, 136 ms, 97 μs, and 3 hnsecs From, Vino.BOn Friday, 6 November 2020 at 04:58:05 UTC, Vino wrote:Hi Andre, The benchmark was performed with 100 executions as below void main() { void f() { Array!string[string] data = getdata("TEST"); foreach(i; data) { writeln(i[]); } } auto r = benchmark!(f)(100); Duration t = r[0]; writeln(t); }[...]While doing performance measurements you should do each test multiple time (at least 5 times). There could be for example an effect that executing a db query the first time is a lot slower than afterwards. Maybe mysql native is slower, but maybe this isn't the case, just the way performance measurements was done was incorrectly. Kind regards Andre
Nov 07 2020
On Saturday, 7 November 2020 at 14:57:39 UTC, Vino wrote:After further analysis we suspect that the issue is at the package std.net.curl the flow of the program is as belowEstablishing a new connection may vary in duration of +200ms or more.
Nov 07 2020
On Saturday, 7 November 2020 at 15:26:48 UTC, frame wrote:On Saturday, 7 November 2020 at 14:57:39 UTC, Vino wrote:Hi All, Trying to improve the above code hence request your help on how to use array container instead of array, tried as per the example below but not working. Code: ubyte[] content; https.onReceive = (ubyte[] data) { content ~= data; return data.length; }; Example Array!ubyte content; https.onReceive = (Array!ubyte data) { content.insert(data); return data.length; }; From, Vino.BAfter further analysis we suspect that the issue is at the package std.net.curl the flow of the program is as belowEstablishing a new connection may vary in duration of +200ms or more.
Nov 07 2020
On Saturday, 7 November 2020 at 16:37:22 UTC, Vino wrote: ,Trying to improve the above code hence request your help on how to use array container instead of array, tried as per the example below but not working.I think you can just use a static array as ubyte[1024] if you want a fixed buffer - it should convert to the ubyte[] slice and compile. Try caching your cURL instance if possible since allocating cURL resources have an impact on overall duration.
Nov 07 2020
On Saturday, 7 November 2020 at 12:29:46 UTC, Andre Pany wrote:Maybe mysql native is slower, but maybe this isn't the case, just the way performance measurements was done was incorrectly.I have tried mysql-native and it tooks way to long for simple things. My test setup is a remote linux vm with a local database and a windows machine with a SSH tunnel connected to that linux database (TCP) I have tested a very simple SELECT query: windows (remote): connect: 50ms execute: 25ms ------------- normal since network latency of ~25ms linux (local): connect: 1ms execute: 44ms (should be <1ms) ------------- abnormal And if you assume network problems: mysql client -h127.0.0.1 says 1 row in set (0.00 sec) as expected Something wrong with this library.
Nov 07 2020
On Sunday, 8 November 2020 at 05:07:54 UTC, frame wrote:Something wrong with this library.Well, we need an edit function here. The library has no TCP_NODELAY flag enabled. This fix speeds it up to 10x and solves the 44ms problem to 400us https://github.com/mysql-d/mysql-native/pull/221/files
Nov 07 2020