Will switching to Java 21's "Virtual Thread" have any benefit over VertX?

[nqminhuit]

• https://openjdk.org/jeps/444
• https://docs.oracle.com/en/java/javase/21/core/virtual-threads.html#GUID-2DDA5807-5BD5-4ABC-B62A-A1230F0566E0

[nqminhuit]

Virtual Threads (VT) are only strong at thousands of concurrent request, our gis does not use that much of IO requests. One benefit of VT I can see now is that it provides the sequential code order, which is handy when we need to print status to stdout. Using asynchronous from Vertx is complex but it make the program lighter and still fast, however, it's difficult to write unit tests.

VT Pros:

• code run in sequential => easier to reason
• built-in java core
• easier to test

VT Cons:

• resource consumption ???
• not too many IO operations to facilitate

Vertx Pros:

• lightweight
• fast

Vertx Cons:

• complex, difficult to reason about
• hard to test
• extra dependencies

We to test performance of VT and Vertx first, if VT faster or similar to Vertx, then it's worth the switch. Otherwise we need to evaluate more. For that to work, we need to finish these 2 issues first: #13 and #15

[nqminhuit]

Performance: 1.1.4 with VertX:

| Command          | Mean [ms]       | Min [ms] | Max [ms] | Relative |
|------------------+-----------------+----------+----------+----------|
| gis              | 98.4 ± 5.9      |     89.0 |    110.9 |     1.00 |
| gis branches -nn | 73.4 ± 3.1      |     65.6 |     82.8 |     1.00 |
| gis branches     | 74.2 ± 3.0      |     66.3 |     79.2 |     1.00 |
| gis co br_tst_1  | 4344.0 ± 2407.7 |   1264.6 |   7656.9 |     1.00 |
| gis fetch        | 5095.3 ± 1722.7 |   3494.3 |   8454.7 |     1.00 |
| gis files        | 3578.6 ± 50.8   |   3535.7 |   3715.6 |     1.00 |
| gis status       | 98.6 ± 4.6      |     88.8 |    107.8 |     1.00 |

2.0.0-dev with Virtual Threads:

| Command          | Mean [ms]       | Min [ms] | Max [ms] | Relative |
|------------------+-----------------+----------+----------+----------|
| gis              | 150.6 ± 18.0    |    114.0 |    175.9 |     1.00 |
| gis branches -nn | 151.3 ± 26.9    |     73.3 |    190.1 |     1.00 |
| gis branches     | 147.8 ± 30.1    |     82.5 |    204.5 |     1.00 |
| gis co br_tst_1  | 2478.8 ± 2192.4 |   1096.8 |   8047.5 |     1.00 |
| gis fetch        | 5665.2 ± 1504.6 |   4200.7 |   8196.0 |     1.00 |
| gis files        | 3470.3 ± 45.2   |   3412.6 |   3556.2 |     1.00 |
| gis status       | 161.6 ± 17.2    |    123.1 |    190.5 |     1.00 |

• Virtual Threads is slightly slower than VertX, however, gis co br_tst_1 is significantly faster and gis files is slighly faster.
• The binary built by Virtual Thread is only 18MB while with VertX it is 40MB
• Code coverage of Virtual Thread is increase from ~25% to 71% and still can be improve.

[nqminhuit]

With Virtual threads we can write maintanable code and facilitate built-in java without extra dependencies. It's worth the switch.