a73x

471ce90d

feat(agent): a lost guest reports what its hypervisor said

a73x   2026-08-06 09:12

Commit message
feat(agent): a lost guest reports what its hypervisor said

An ephemeral VM that dies is reported "ephemeral VM lost" and nothing else. That
is a true statement about the process table and a useless one to debug from: it
names neither what failed nor why, and it is the only report such a VM ever
produces, because ephemeral VMs are never restarted. A guest handed an image its
host cannot execute dies exactly here — no serial output, no address, no clue.

The cause was already on disk. Every backend redirects its hypervisor's stdout
and stderr to a file beside the VM's disk — cloud-hypervisor to ch.log, vfkit to
vfkit.log — and nothing ever read them. internal/agent/hyperlog quotes the last
meaningful line back, and the lost report carries it.

It is the hypervisor's own words, not an interpretation of them. eitri cannot
enumerate what cloud-hypervisor or vfkit might fail at, and a guess dressed up as
a diagnosis is worse than a quote.

FailureReason joins the Provisioner seam rather than arriving as an optional
interface a caller has to test for: both backends keep such a log, so a backend
that cannot answer is a backend that says so by returning empty. Empty means
"nothing to add", never "nothing went wrong" — a guest killed by a host reboot
leaves no complaint behind and is still lost.

The quote is bounded on both sides of the read: only the tail of the file is
examined, so a log running for weeks is never pulled into memory, and the line
is stripped of control characters and capped before it reaches last_error, which
is a database column rendered in the console and the CLI.
docs/shape.html
Old New
@@ -146,6 +146,7 @@
146 "imports": [ 146 "imports": [
147 "internal/agent/exec", 147 "internal/agent/exec",
148 "internal/agent/hostinfo", 148 "internal/agent/hostinfo",
149 "internal/agent/hyperlog",
149 "internal/agent/pidfile", 150 "internal/agent/pidfile",
150 "internal/agent/state" 151 "internal/agent/state"
151 ] 152 ]
@@ -180,6 +181,12 @@
180 ] 181 ]
181 }, 182 },
182 { 183 {
184 "importPath": "internal/agent/hyperlog",
185 "plane": "data",
186 "synopsis": "Package hyperlog reads back the last thing a hypervisor said before it stopped running.",
187 "imports": []
188 },
189 {
183 "importPath": "internal/agent/imagecache", 190 "importPath": "internal/agent/imagecache",
184 "plane": "data", 191 "plane": "data",
185 "synopsis": "Package imagecache downloads and verifies content-addressed base images (decoded to raw in-process, LRU-evicted beyond MaxBytes).", 192 "synopsis": "Package imagecache downloads and verifies content-addressed base images (decoded to raw in-process, LRU-evicted beyond MaxBytes).",
@@ -283,6 +290,7 @@
283 "imports": [ 290 "imports": [
284 "internal/agent/exec", 291 "internal/agent/exec",
285 "internal/agent/hostinfo", 292 "internal/agent/hostinfo",
293 "internal/agent/hyperlog",
286 "internal/agent/pidfile", 294 "internal/agent/pidfile",
287 "internal/agent/state" 295 "internal/agent/state"
288 ] 296 ]
docs/shape.json
Old New
@@ -95,6 +95,7 @@
95 "imports": [ 95 "imports": [
96 "internal/agent/exec", 96 "internal/agent/exec",
97 "internal/agent/hostinfo", 97 "internal/agent/hostinfo",
98 "internal/agent/hyperlog",
98 "internal/agent/pidfile", 99 "internal/agent/pidfile",
99 "internal/agent/state" 100 "internal/agent/state"
100 ] 101 ]
@@ -129,6 +130,12 @@
129 ] 130 ]
130 }, 131 },
131 { 132 {
133 "importPath": "internal/agent/hyperlog",
134 "plane": "data",
135 "synopsis": "Package hyperlog reads back the last thing a hypervisor said before it stopped running.",
136 "imports": []
137 },
138 {
132 "importPath": "internal/agent/imagecache", 139 "importPath": "internal/agent/imagecache",
133 "plane": "data", 140 "plane": "data",
134 "synopsis": "Package imagecache downloads and verifies content-addressed base images (decoded to raw in-process, LRU-evicted beyond MaxBytes).", 141 "synopsis": "Package imagecache downloads and verifies content-addressed base images (decoded to raw in-process, LRU-evicted beyond MaxBytes).",
@@ -232,6 +239,7 @@
232 "imports": [ 239 "imports": [
233 "internal/agent/exec", 240 "internal/agent/exec",
234 "internal/agent/hostinfo", 241 "internal/agent/hostinfo",
242 "internal/agent/hyperlog",
235 "internal/agent/pidfile", 243 "internal/agent/pidfile",
236 "internal/agent/state" 244 "internal/agent/state"
237 ] 245 ]
internal/agent/cloudhv/cloudhv.go
Old New
@@ -18,6 +18,7 @@ import (
18 18
19 agentexec "github.com/a73x/eitri/internal/agent/exec" 19 agentexec "github.com/a73x/eitri/internal/agent/exec"
20 "github.com/a73x/eitri/internal/agent/hostinfo" 20 "github.com/a73x/eitri/internal/agent/hostinfo"
21 "github.com/a73x/eitri/internal/agent/hyperlog"
21 "github.com/a73x/eitri/internal/agent/pidfile" 22 "github.com/a73x/eitri/internal/agent/pidfile"
22 "github.com/a73x/eitri/internal/agent/state" 23 "github.com/a73x/eitri/internal/agent/state"
23 ) 24 )
@@ -269,8 +270,7 @@ func (p *Provisioner) Boot(ctx context.Context, vmID string, spec state.VMSpec)
269 // Guest console output (serial) goes to a unix socket (--serial socket=…) 270 // Guest console output (serial) goes to a unix socket (--serial socket=…)
270 // that the serialpump drains into serial.log. 271 // that the serialpump drains into serial.log.
271 // CH's own diagnostic output (startup errors, API logs) goes to ch.log. 272 // CH's own diagnostic output (startup errors, API logs) goes to ch.log.
272 chLogPath := filepath.Join(p.st.VMDir(vmID), "ch.log") 273 chLog, err := os.OpenFile(p.logPath(vmID), os.O_CREATE|os.O_APPEND|os.O_WRONLY, chLogMode)
273 chLog, err := os.OpenFile(chLogPath, os.O_CREATE|os.O_APPEND|os.O_WRONLY, chLogMode)
274 if err != nil { 274 if err != nil {
275 return fmt.Errorf("open ch.log %s: %w", vmID, err) 275 return fmt.Errorf("open ch.log %s: %w", vmID, err)
276 } 276 }
@@ -340,6 +340,17 @@ func (p *Provisioner) ownedPID(vmID string) int {
340 // agent wrote cannot have been recycled while its process is alive. reconcile's 340 // agent wrote cannot have been recycled while its process is alive. reconcile's
341 // own boot-ID check still decides what a lost VM means — this only decides 341 // own boot-ID check still decides what a lost VM means — this only decides
342 // which process, if any, is ours to ask about. 342 // which process, if any, is ours to ask about.
343 // logPath is where cloud-hypervisor's own diagnostic output is kept (startup
344 // errors, API logs) — distinct from serial.log, which carries the GUEST's
345 // console.
346 func (p *Provisioner) logPath(vmID string) string {
347 return filepath.Join(p.st.VMDir(vmID), "ch.log")
348 }
349
350 // FailureReason quotes the last thing cloud-hypervisor said. See
351 // reconcile.Provisioner.
352 func (p *Provisioner) FailureReason(vmID string) string { return hyperlog.Reason(p.logPath(vmID)) }
353
343 func (p *Provisioner) Running(vmID string) bool { 354 func (p *Provisioner) Running(vmID string) bool {
344 pid := p.ownedPID(vmID) 355 pid := p.ownedPID(vmID)
345 if pid == 0 { 356 if pid == 0 {
internal/agent/cloudhv/cloudhv_test.go
Old New
@@ -666,3 +666,20 @@ func TestAPidfileWithoutABootIDIsStillOurs(t *testing.T) {
666 666
667 assert.True(t, p.Running(vmID), "an agent upgrade must not orphan the guests it inherits") 667 assert.True(t, p.Running(vmID), "an agent upgrade must not orphan the guests it inherits")
668 } 668 }
669
670 // TestFailureReasonQuotesCHLog pins the wiring, not the tail logic (that is
671 // hyperlog's own test): FailureReason must read the SAME file Boot redirects
672 // cloud-hypervisor's stdout and stderr to. A backend quoting the wrong file
673 // reports nothing forever and looks exactly like a backend with nothing to say.
674 func TestFailureReasonQuotesCHLog(t *testing.T) {
675 st, err := state.Open(t.TempDir())
676 require.NoError(t, err)
677 p := New(st, "ch", "fw", nil, newFakeNet())
678
679 require.Empty(t, p.FailureReason("vm1"), "no log yet means nothing to add")
680
681 require.NoError(t, os.MkdirAll(st.VMDir("vm1"), 0o700))
682 require.NoError(t, os.WriteFile(filepath.Join(st.VMDir("vm1"), "ch.log"),
683 []byte("cloud-hypervisor booting\nError: VmBoot(DeviceManager(Kernel))\n"), 0o600))
684 assert.Equal(t, "Error: VmBoot(DeviceManager(Kernel))", p.FailureReason("vm1"))
685 }
internal/agent/hyperlog/hyperlog.go
Old New
@@ -0,0 +1,117 @@
1 // Package hyperlog reads back the last thing a hypervisor said before it
2 // stopped running.
3 //
4 // Every backend already redirects its hypervisor's stdout and stderr to a file
5 // beside the VM's disk — cloud-hypervisor to ch.log, vfkit to vfkit.log — but
6 // nothing ever read them, so a guest whose hypervisor exited on startup was
7 // reported as lost and nothing more. "ephemeral VM lost" is a true statement
8 // about the process table and says nothing about the cause; the cause was
9 // sitting on disk the whole time.
10 //
11 // This is deliberately the hypervisor's own words rather than an interpretation
12 // of them. eitri cannot enumerate what cloud-hypervisor or vfkit might fail at,
13 // and a guess dressed up as a diagnosis is worse than a quote.
14 package hyperlog
15
16 import (
17 "bytes"
18 "io"
19 "os"
20 "strings"
21 "unicode"
22 )
23
24 const (
25 // tailBytes bounds the read. A hypervisor's fatal message is its last line,
26 // so only the end of the file is interesting, and a log that has been
27 // running for weeks must not be pulled into memory to find it.
28 tailBytes = 8 << 10
29 // maxLen bounds the returned string. It lands in a VM's last_error, which
30 // is a database column rendered in the console and the CLI, so a runaway
31 // line has to be cut somewhere it stays readable.
32 maxLen = 240
33 )
34
35 // Reason returns the last meaningful line the hypervisor wrote to path, or ""
36 // when there is nothing to report — no file, an empty one, or output that is
37 // all whitespace. Empty means "nothing to add", never "nothing went wrong":
38 // callers append it to their own message only when it is non-empty.
39 func Reason(path string) string {
40 f, err := os.Open(path)
41 if err != nil {
42 return ""
43 }
44 defer f.Close()
45
46 // Seek to the last tailBytes. A file shorter than that is read whole; the
47 // first line of the window may be a fragment, which is why only complete
48 // trailing lines are considered below.
49 size, err := f.Seek(0, io.SeekEnd)
50 if err != nil {
51 return ""
52 }
53 start := max(size-tailBytes, 0)
54 if _, err := f.Seek(start, io.SeekStart); err != nil {
55 return ""
56 }
57 buf, err := io.ReadAll(f)
58 if err != nil {
59 return ""
60 }
61 // Drop a leading partial line when the window did not start at the file's
62 // beginning, so a truncated fragment is never reported as the reason.
63 if start > 0 {
64 if i := bytes.IndexByte(buf, '\n'); i >= 0 {
65 buf = buf[i+1:]
66 } else {
67 return "" // one enormous line, no complete one to quote
68 }
69 }
70
71 lines := strings.Split(string(buf), "\n")
72 for i := len(lines) - 1; i >= 0; i-- {
73 if s := clean(lines[i]); s != "" {
74 return s
75 }
76 }
77 return ""
78 }
79
80 // clean reduces one log line to something safe to store and show: printable
81 // characters only, collapsed whitespace, bounded length. Hypervisors colour
82 // their output and draw progress with control characters, none of which belong
83 // in a database column.
84 func clean(line string) string {
85 var b strings.Builder
86 b.Grow(len(line))
87 prevSpace := true // leading whitespace is dropped
88 for _, r := range line {
89 switch {
90 case r == 0x1b: // start of an ANSI escape; drop the rest of the line
91 return strings.TrimRight(b.String(), " ")
92 case unicode.IsSpace(r):
93 if !prevSpace {
94 b.WriteRune(' ')
95 prevSpace = true
96 }
97 case unicode.IsPrint(r):
98 b.WriteRune(r)
99 prevSpace = false
100 }
101 // Anything else (other control characters) is dropped.
102 }
103 out := strings.TrimRight(b.String(), " ")
104 if len(out) > maxLen {
105 // Cut on a rune boundary so the result stays valid UTF-8.
106 cut := maxLen
107 for cut > 0 && !utf8Start(out[cut]) {
108 cut--
109 }
110 out = strings.TrimRight(out[:cut], " ") + "…"
111 }
112 return out
113 }
114
115 // utf8Start reports whether b begins a UTF-8 rune (i.e. is not a continuation
116 // byte).
117 func utf8Start(b byte) bool { return b&0xC0 != 0x80 }
internal/agent/hyperlog/hyperlog_test.go
Old New
@@ -0,0 +1,84 @@
1 package hyperlog
2
3 import (
4 "os"
5 "path/filepath"
6 "strings"
7 "testing"
8 )
9
10 // write puts body in a temp file and returns its path.
11 func write(t *testing.T, body string) string {
12 t.Helper()
13 p := filepath.Join(t.TempDir(), "hv.log")
14 if err := os.WriteFile(p, []byte(body), 0o600); err != nil {
15 t.Fatal(err)
16 }
17 return p
18 }
19
20 func TestReason(t *testing.T) {
21 for _, tc := range []struct {
22 name, body, want string
23 }{
24 {"last line wins", "starting\nvcpu 0 ready\nError: VmBoot(DeviceManager)\n",
25 "Error: VmBoot(DeviceManager)"},
26 {"trailing blank lines skipped", "Error: no bootable device\n\n\n \n",
27 "Error: no bootable device"},
28 {"no trailing newline", "line one\nfatal: bad kernel", "fatal: bad kernel"},
29 {"single line", "vfkit: failed to start", "vfkit: failed to start"},
30 {"empty file", "", ""},
31 {"whitespace only", "\n\n \t\n", ""},
32 {"collapses whitespace", "Error: too many\tspaces\n", "Error: too many spaces"},
33 {"strips control characters", "Error: bad\x00 image\x07\n", "Error: bad image"},
34 {"truncates at an ANSI escape", "\x1b[31mError: red\x1b[0m\n", ""},
35 } {
36 t.Run(tc.name, func(t *testing.T) {
37 if got := Reason(write(t, tc.body)); got != tc.want {
38 t.Errorf("Reason() = %q, want %q", got, tc.want)
39 }
40 })
41 }
42 }
43
44 // TestReasonMissingFile pins that an absent log is "nothing to add" rather than
45 // an error the caller has to handle: a VM that never started writes no log, and
46 // that is not itself a fault worth reporting.
47 func TestReasonMissingFile(t *testing.T) {
48 if got := Reason(filepath.Join(t.TempDir(), "absent.log")); got != "" {
49 t.Errorf("Reason() on a missing file = %q, want empty", got)
50 }
51 if got := Reason(t.TempDir()); got != "" {
52 t.Errorf("Reason() on a directory = %q, want empty", got)
53 }
54 }
55
56 // TestReasonReadsOnlyTheTail pins that a long-lived log is not pulled into
57 // memory whole, and that the partial first line of the read window is never
58 // mistaken for the reason.
59 func TestReasonReadsOnlyTheTail(t *testing.T) {
60 body := strings.Repeat("chatter chatter chatter\n", 4000) + "Error: the last word\n"
61 if got := Reason(write(t, body)); got != "Error: the last word" {
62 t.Errorf("Reason() = %q, want the final line", got)
63 }
64 // A window that lands mid-line must not quote the fragment: a file that is
65 // one enormous line has no complete trailing line to report.
66 if got := Reason(write(t, strings.Repeat("x", 40<<10))); got != "" {
67 t.Errorf("Reason() on one huge line = %q, want empty", got)
68 }
69 }
70
71 // TestReasonBoundsLength pins the cap: last_error is a database column shown in
72 // the console, so a runaway line is cut rather than stored whole.
73 func TestReasonBoundsLength(t *testing.T) {
74 got := Reason(write(t, "Error: "+strings.Repeat("verbose ", 200)+"\n"))
75 if len(got) > maxLen+len("…") {
76 t.Errorf("Reason() len = %d, want <= %d", len(got), maxLen+len("…"))
77 }
78 if !strings.HasSuffix(got, "…") {
79 t.Errorf("a truncated reason must say so, got %q", got)
80 }
81 if !strings.HasPrefix(got, "Error: verbose") {
82 t.Errorf("truncation must keep the START of the line, got %q", got)
83 }
84 }
internal/agent/reconcile/reconcile.go
Old New
@@ -92,6 +92,16 @@ type Provisioner interface {
92 92
93 Running(vmID string) bool 93 Running(vmID string) bool
94 94
95 // FailureReason returns what the hypervisor said before it stopped running,
96 // or "" when the backend has nothing to add. It is asked only once a VM has
97 // been found lost, to give that report a cause: the process table can say a
98 // guest is gone but never why, and every backend already keeps its
99 // hypervisor's own output on disk.
100 //
101 // Empty means "nothing to add", NOT "nothing went wrong" — a guest killed
102 // by a host reboot leaves no complaint behind and is still lost.
103 FailureReason(vmID string) string
104
95 // Address returns the VM's current guest address, or "" when the backend 105 // Address returns the VM's current guest address, or "" when the backend
96 // does not know one (never booted, or gone). It is POLLED rather than 106 // does not know one (never booted, or gone). It is POLLED rather than
97 // returned by Boot: where the host OS's own DHCP server hands out the 107 // returned by Boot: where the host OS's own DHCP server hands out the
@@ -808,8 +818,15 @@ func (e *Engine) converge(ctx context.Context, d *pb.VMDesired, rec state.Record
808 818
809 if lost { 819 if lost {
810 if !d.Persistent { 820 if !d.Persistent {
811 // Ephemeral lost VMs are reported failed and NEVER restarted. 821 // Ephemeral lost VMs are reported failed and NEVER restarted. This
822 // is the only report the operator gets, so it carries whatever the
823 // hypervisor said on its way out — a guest handed an image its host
824 // cannot execute dies here, and "ephemeral VM lost" alone names
825 // neither the image nor the architecture.
812 errMsg := "ephemeral VM lost" 826 errMsg := "ephemeral VM lost"
827 if why := e.Prov.FailureReason(d.VmId); why != "" {
828 errMsg += ": " + why
829 }
813 rec.LastError = errMsg 830 rec.LastError = errMsg
814 _ = e.St.SaveVM(rec) 831 _ = e.St.SaveVM(rec)
815 res.report(d.VmId, rec.IP, "stopped", "failed", errMsg) 832 res.report(d.VmId, rec.IP, "stopped", "failed", errMsg)
internal/agent/reconcile/reconcile_test.go
Old New
@@ -29,13 +29,13 @@ import (
29 // addressing is only observable through the seam, so the fake has to own it for 29 // addressing is only observable through the seam, so the fake has to own it for
30 // these tests to mean anything. 30 // these tests to mean anything.
31 type fakeProv struct { 31 type fakeProv struct {
32 mu sync.Mutex 32 mu sync.Mutex
33 running map[string]bool 33 running map[string]bool
34 prepCalls int // total PrepareRootDisk invocations, including failed ones 34 prepCalls int // total PrepareRootDisk invocations, including failed ones
35 prepared []string 35 prepared []string
36 booted []string 36 booted []string
37 shutdown []string 37 shutdown []string
38 destroyed []string 38 destroyed []string
39 prepErr error 39 prepErr error
40 bootErr error // one-shot: consumed and cleared on first Boot call 40 bootErr error // one-shot: consumed and cleared on first Boot call
41 destroyErr error // sticky: every Destroy fails until it is cleared 41 destroyErr error // sticky: every Destroy fails until it is cleared
@@ -44,6 +44,10 @@ type fakeProv struct {
44 cidr string 44 cidr string
45 addrs map[string]string // vmID -> ip (sticky, mirrors the DHCP table) 45 addrs map[string]string // vmID -> ip (sticky, mirrors the DHCP table)
46 46
47 // failReason is what the fake's "hypervisor" left behind; empty is a
48 // backend with nothing to add, which is the common case.
49 failReason string
50
47 // lateAddress switches the fake to the shape a host-run DHCP server has: 51 // lateAddress switches the fake to the shape a host-run DHCP server has:
48 // Boot attaches the NIC but assigns nothing, and the address exists only 52 // Boot attaches the NIC but assigns nothing, and the address exists only
49 // once the guest has booted and asked for one (answerAddress). 53 // once the guest has booted and asked for one (answerAddress).
@@ -163,6 +167,12 @@ func (f *fakeProv) Running(id string) bool {
163 return f.running[id] 167 return f.running[id]
164 } 168 }
165 169
170 func (f *fakeProv) FailureReason(string) string {
171 f.mu.Lock()
172 defer f.mu.Unlock()
173 return f.failReason
174 }
175
166 type fixture struct { 176 type fixture struct {
167 eng *Engine 177 eng *Engine
168 prov *fakeProv 178 prov *fakeProv
@@ -389,6 +399,36 @@ func TestUserStopIsStoppedNotLost(t *testing.T) {
389 assert.NotEqual(t, "failed", av.Phase, "recorded stop request: stopped != lost") 399 assert.NotEqual(t, "failed", av.Phase, "recorded stop request: stopped != lost")
390 } 400 }
391 401
402 // TestEphemeralLostCarriesTheHypervisorsReason pins that the one report an
403 // ephemeral VM ever produces says WHY. "ephemeral VM lost" is a true statement
404 // about the process table and a useless one to debug from: a guest handed an
405 // image its host cannot execute dies exactly here, and the hypervisor's own
406 // complaint is the only thing that names the cause.
407 func TestEphemeralLostCarriesTheHypervisorsReason(t *testing.T) {
408 f := setup(t)
409 f.step(snap(1, vm("vm1")))
410 f.prov.running["vm1"] = false
411 f.prov.failReason = "Error: VmBoot(NoBootableDevice)"
412
413 av := findVM(f.step(snap(1, vm("vm1"))), "vm1")
414 assert.Equal(t, "failed", av.Phase)
415 assert.Equal(t, "ephemeral VM lost: Error: VmBoot(NoBootableDevice)", av.LastError)
416 }
417
418 // TestEphemeralLostWithoutAReasonStaysBare pins the empty case: a backend with
419 // nothing to add must not produce a dangling separator. Empty means "nothing to
420 // add", not "nothing went wrong" — a guest killed by a host reboot leaves no
421 // complaint behind and is still lost.
422 func TestEphemeralLostWithoutAReasonStaysBare(t *testing.T) {
423 f := setup(t)
424 f.step(snap(1, vm("vm1")))
425 f.prov.running["vm1"] = false
426 f.prov.failReason = ""
427
428 av := findVM(f.step(snap(1, vm("vm1"))), "vm1")
429 assert.Equal(t, "ephemeral VM lost", av.LastError)
430 }
431
392 func TestEphemeralLostOnHostRebootNeverRestarts(t *testing.T) { 432 func TestEphemeralLostOnHostRebootNeverRestarts(t *testing.T) {
393 f := setup(t) 433 f := setup(t)
394 f.step(snap(1, vm("vm1"))) 434 f.step(snap(1, vm("vm1")))
internal/agent/syncclient/client_test.go
Old New
@@ -37,6 +37,7 @@ func (noopProv) Shutdown(context.Context, string) error { r
37 func (noopProv) Destroy(context.Context, string) error { return nil } 37 func (noopProv) Destroy(context.Context, string) error { return nil }
38 func (noopProv) Running(string) bool { return false } 38 func (noopProv) Running(string) bool { return false }
39 func (noopProv) Address(string) string { return "10.77.1.2" } 39 func (noopProv) Address(string) string { return "10.77.1.2" }
40 func (noopProv) FailureReason(string) string { return "" }
40 41
41 // testQUICIdle is the deliberately-short idle timeout the test listeners use so 42 // testQUICIdle is the deliberately-short idle timeout the test listeners use so
42 // reconnect/drop cases don't wait out the production SyncMaxIdleTimeout. 43 // reconnect/drop cases don't wait out the production SyncMaxIdleTimeout.
internal/agent/vfkit/vfkit.go
Old New
@@ -32,6 +32,7 @@ import (
32 32
33 agentexec "github.com/a73x/eitri/internal/agent/exec" 33 agentexec "github.com/a73x/eitri/internal/agent/exec"
34 "github.com/a73x/eitri/internal/agent/hostinfo" 34 "github.com/a73x/eitri/internal/agent/hostinfo"
35 "github.com/a73x/eitri/internal/agent/hyperlog"
35 "github.com/a73x/eitri/internal/agent/pidfile" 36 "github.com/a73x/eitri/internal/agent/pidfile"
36 "github.com/a73x/eitri/internal/agent/state" 37 "github.com/a73x/eitri/internal/agent/state"
37 ) 38 )
@@ -176,6 +177,10 @@ func (p *Provisioner) sockPath(vmID string) string { return p.vmFile(vmID, "vfki
176 func (p *Provisioner) varStore(vmID string) string { return p.vmFile(vmID, "efi-vars.fd") } 177 func (p *Provisioner) varStore(vmID string) string { return p.vmFile(vmID, "efi-vars.fd") }
177 func (p *Provisioner) logPath(vmID string) string { return p.vmFile(vmID, "vfkit.log") } 178 func (p *Provisioner) logPath(vmID string) string { return p.vmFile(vmID, "vfkit.log") }
178 179
180 // FailureReason quotes the last thing vfkit said. See
181 // reconcile.Provisioner.
182 func (p *Provisioner) FailureReason(vmID string) string { return hyperlog.Reason(p.logPath(vmID)) }
183
179 // SocketPath is the VM's vfkit REST socket, exported so the composition root 184 // SocketPath is the VM's vfkit REST socket, exported so the composition root
180 // can hand ConsoleSource the same path Boot writes. 185 // can hand ConsoleSource the same path Boot writes.
181 func (p *Provisioner) SocketPath(vmID string) string { return p.sockPath(vmID) } 186 func (p *Provisioner) SocketPath(vmID string) string { return p.sockPath(vmID) }
internal/agent/vfkit/vfkit_test.go
Old New
@@ -653,3 +653,17 @@ func TestShutdownDoesNotLeakAConnectionPerCall(t *testing.T) {
653 } 653 }
654 }) 654 })
655 } 655 }
656
657 // TestFailureReasonQuotesVfkitLog pins the wiring, not the tail logic (that is
658 // hyperlog's own test): FailureReason must read the SAME file Boot redirects
659 // vfkit's stdout and stderr to.
660 func TestFailureReasonQuotesVfkitLog(t *testing.T) {
661 p := newTestProv(t, nil)
662
663 require.Empty(t, p.FailureReason("vm1"), "no log yet means nothing to add")
664
665 require.NoError(t, os.MkdirAll(p.st.VMDir("vm1"), 0o700))
666 require.NoError(t, os.WriteFile(p.logPath("vm1"),
667 []byte("vfkit starting\nvirtual machine failed to start: unsupported guest\n"), 0o600))
668 assert.Equal(t, "virtual machine failed to start: unsupported guest", p.FailureReason("vm1"))
669 }
scripts/coverage.sh
Old New
@@ -24,6 +24,7 @@ declare -A FLOOR=(
24 [internal/agent/imagecache]=72 24 [internal/agent/imagecache]=72
25 [internal/agent/netenv]=76 25 [internal/agent/netenv]=76
26 [internal/agent/cloudhv]=80 26 [internal/agent/cloudhv]=80
27 [internal/agent/hyperlog]=90
27 [internal/agent/pidfile]=95 28 [internal/agent/pidfile]=95
28 [internal/agent/vfkit]=85 29 [internal/agent/vfkit]=85
29 [internal/agent/syncclient]=74 30 [internal/agent/syncclient]=74