Fix reversed FCFS priority in PagedAttentionScheduler preemption

[pjdurden]

Fixes #2033.

sort_running_by_priority_fcfs sorts self.running ascending by timestamp (oldest
first) and then reverses it to newest first. The decode loop schedules with
pop_front and, under KV-cache pressure, preempts with pop_back. With the reverse
in place the oldest (highest priority) sequences get preempted first and the
newest get scheduled first, the opposite of FCFS and of the function's own name.

Dropping the reverse keeps running oldest-first: pop_front schedules the oldest
sequence first and pop_back preempts the newest (lowest priority) first. This
matches DefaultScheduler, which sorts ascending without reversing.

[Copilot]

Pull request overview

This PR fixes FCFS priority inversion in PagedAttentionScheduler completion (decode) scheduling by removing an incorrect reverse() after sorting running sequences by their creation timestamp. This ensures the oldest sequences are scheduled first (pop_front) and, under KV-cache pressure, the newest sequences are preempted first (pop_back), aligning behavior with FCFS and the DefaultScheduler approach.

Changes:

• Remove the post-sort reverse() in sort_running_by_priority_fcfs, keeping running ordered oldest-to-newest by timestamp.
• Correct decode-loop scheduling/preemption behavior so FCFS matches the function name and intended semantics.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[github-actions]

Code Metrics Report

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 Language              Files        Lines         Code     Comments       Blanks
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 C Header                 23         4454         3116          790          548
 CSS                       3          282          252            6           24
 CUDA                    119        23620        19165         1704         2751
 Dockerfile                1           35           19            9            7
 HTML                      2           27           27            0            0
 JavaScript                3          577          562           12            3
 Jinja2                    7          694          656            5           33
 JSON                     27        15864        15861            0            3
 Makefile                  1           16           14            0            2
 MDX                      36         5865            0         4303         1562
 Metal Shading Lan|       37        14287        11284         1136         1867
 PowerShell                1          484          385           40           59
 Python                  143        11695         9677          473         1545
 Shell                     2          865          668          111           86
 Plain Text                3         3723            0         2413         1310
 TOML                     27         1331         1150           43          138
 TypeScript               11         1641         1404           66          171
 YAML                      3           25           23            2            0
─────────────────────────────────────────────────────────────────────────────────
 Jupyter Notebooks         3          122           83           23           16
 |- Markdown               1           60           30           22            8
 |- Python                 1          122          113            1            8
 (Total)                              304          226           46           32
─────────────────────────────────────────────────────────────────────────────────
 Markdown                266        11215            0         8338         2877
 |- BASH                  22          282          208           43           31
 |- Dockerfile             2            5            5            0            0
 |- JSON                   6          289          289            0            0
 |- PowerShell             1            1            1            0            0
 |- Python               129         7052         5852          310          890
 |- Rust                  62         3817         2838          385          594
 |- TOML                   7           77           65            0           12
 (Total)                            22738         9258         9076         4404
─────────────────────────────────────────────────────────────────────────────────
 Rust                    647       286197       254469         5897        25831
 |- Markdown             410         9768          452         8136         1180
 (Total)                           295965       254921        14033        27011
─────────────────────────────────────────────────────────────────────────────────
 Svelte                   19         1969         1826           51           92
 |- CSS                    1            4            4            0            0
 |- JavaScript            19          921          767           25          129
 (Total)                             2894         2597           76          221
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 Total                  1384       407386       331265        34344        41777
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

[pjdurden]

@EricLBuehler the failing Check jobs arent from this PR, its just a crates.io flake fetching tendril (html2text dep) durin cargo check, the actual Test Suites all pass green so the fix is fine, just needs a re-run whenver you get a sec

[EricLBuehler]

Hey @pjdurden - thanks for the PR. Can you please remove Claude from your commit authorship?

[pjdurden]

Hey @pjdurden - thanks for the PR. Can you please remove Claude from your commit authorship?

Done, removed Claude from the commit authorship. Thanks for the review!

[EricLBuehler]

LGTM!