myxce/src/frontend/docs/improvement_suggestions.md

Code Improvement Suggestions

Date: 2026-04-09
Based on current codebase state after all bug fixes.

---

1. Critical Bugs Still Present

1.1 queue_name filter added N times in GetJobsInfo (apiserver)

File: controllers/jobs.go ~line 226
Problem: The queue_name:in:(...) RBAC filter is appended inside the per-filter loop — once for every non-queue filter in the request. With 2 filters (e.g. job_status + submit_user), it gets added twice. Seen in server logs: queue_name:in:(normal,priority,short) appears twice in every query.

Fix: Move the accessQueues append outside the loop, after it completes.

// BEFORE (inside loop — adds queue_name once per non-queue filter)
for i, filter := range filters {
    ...
    } else {
        if len(accessQueues) > 0 {
            filters = append(filters, ...)  // BUG: runs N times
        }
    }
}

// AFTER (outside loop — adds once)
for i, filter := range filters { ... }
if len(accessQueues) > 0 && !queueFilterPresent {
    filters = append(filters, utils.FilterCondition{
        Field: "queue_name", Operator: "in",
        Value: "(" + strings.Join(accessQueues, ",") + ")",
    })
}

---

2. Design Issues

2.1 controllers/jobs.go is 3590 lines — needs splitting

The file handles job query, submit, control, restart, array jobs, and verbose output. Each is a distinct concern.

Suggested split:

controllers/
  jobs_query.go    — GetJobsInfo, GetJobByIDs (~600 lines)
  jobs_submit.go   — CreateJob, RestartJobByID (~500 lines)
  jobs_control.go  — ControlJob, ControlJobByID (~400 lines)
  jobs_array.go    — array job handling (~300 lines)
  jobs_helpers.go  — shared: parseArrayName, processJobData, etc.

2.2 RBAC permission check duplicated across controllers

GetRequireResourcePermission + accessQueues logic is copy-pasted in GetJobsInfo, ControlJob, ControlJobByID, RestartJobByID. Extract to a helper:

// controllers/rbac_helpers.go
func getJobAccessQueues(username string) ([]string, error) { ... }
func checkJobViewPermission(username, queueName string) error { ... }

2.3 xcecli/cmd/jobs/submit.go is 734 lines

JobSubmitter mixes CLI flag parsing, file upload, interactive job handling, and HTTP submission. Split:

cmd/jobs/
  submit.go          — cobra command + flag binding only
  submitter.go       — JobSubmitter struct + BuildJobRequest
  submit_files.go    — file upload logic (already in files.go ✓)
  submit_interactive.go — interactive/TTY/xjob handling

---

3. Code Quality

3.1 ioutil.ReadFile deprecated (Go 1.16)

File: controllers/rbac_check.go:54
Fix: os.ReadFile(path) — one-line change, tracked as Q3.

3.2 Inconsistent error return convention

Some functions return (error, T), others return (T, error). Go convention is always (T, error).

// Non-standard (apiserver/controllers/rbac_check.go)
func GetRequireResourcePermission(...) (error, models.RolePermission)

// Should be
func GetRequireResourcePermission(...) (models.RolePermission, error)

This affects ~8 call sites in jobs.go that do err, perm := GetRequire....

3.3 Magic strings for domain types

"all", "self", "self_ug", "queue" appear as raw strings in multiple files. They're defined as constants in models/roles.go but not always used.

Fix: Replace all raw string comparisons with the constants (models.DomainTypeAll, etc.).

3.4 xcecli output printers have no shared time formatting

page_print_jobs.go now has parseJobTime/formatJobTime. The same pattern will be needed for start_time, end_time in other printers. Move to common/output/time.go.

---

4. Architecture

4.1 In-memory store has no persistence

After restart, all submitted jobs (non-builtin) are lost. The lib/pq PostgreSQL driver is already in go.mod. When backend integration happens, the notification handlers (OnJobCreated, etc.) should write through to PostgreSQL, not just the in-memory store.

4.2 UserRoleSeeder only implemented for config_file

After the startup fix, PAM/LDAP/OAuth2 providers don't implement AllUsers() — so after restart with those providers, users still need to re-login to get their roles. Consider storing user-role assignments in a persistent store (DB or file) so they survive restarts regardless of provider.

4.3 Rate limiter limiters map grows unboundedly

File: middleware/rate_limit.go
The per-endpoint limiters map is never evicted. Under normal operation with fixed routes this is fine, but worth noting for future dynamic routing.

---

5. Priority Order

quadrantChart
    title Impact vs Effort
    x-axis Low Effort --> High Effort
    y-axis Low Impact --> High Impact
    quadrant-1 Do First
    quadrant-2 Plan
    quadrant-3 Later
    quadrant-4 Quick Wins
    queue_name double-add: [0.10, 0.90]
    ioutil fix: [0.05, 0.20]
    error convention: [0.25, 0.60]
    split jobs.go: [0.60, 0.70]
    split submit.go: [0.50, 0.50]
    RBAC helper: [0.35, 0.65]
    time.go shared: [0.15, 0.35]
    persistence: [0.80, 0.85]
    UserRoleSeeder all providers: [0.30, 0.55]

Do immediately:

1. Fix queue_name double-add bug in GetJobsInfo (10 min, high impact)
2. ioutil → os.ReadFile (1 min)

Next sprint: 3. Fix (error, T) → (T, error) return convention 4. Extract RBAC helpers to reduce duplication in jobs.go 5. Split jobs.go into focused files

Backend integration phase: 6. PostgreSQL persistence via notification handlers 7. UserRoleSeeder for PAM/LDAP providers