Resume interrupted downloads instead of restarting from scratch

MSVC/WinSDK/WDK/DXSDK payloads run into the hundreds of MB to several
GB, so a dropped connection or a retry after a transient error used to
mean throwing away everything already fetched and starting over from
byte 0. Track progress in a dest+".part" file and resume it via an
HTTP Range request, falling back to a full restart when the server
doesn't honor Range (200 instead of 206) or the local part is stale
(416).
This commit is contained in:
Cheviiot
2026-07-25 18:17:59 +10:00
parent 98bac75767
commit aa469b329c
2 changed files with 215 additions and 6 deletions
+42 -6
View File
@@ -149,27 +149,63 @@ func tryDownloadPayload(payload Payload, dest, fileID string, allowHashMismatch
var downloadHTTPClient = &http.Client{Timeout: 30 * time.Minute}
// httpDownloadFile downloads url to dest via a dest+".part" temp file,
// resuming from wherever a previous attempt left off if one exists - MSVC/
// WinSDK/WDK/DXSDK payloads run into the hundreds of MB to multiple GB, so
// restarting an interrupted download from byte 0 (a dropped connection, a
// retry after this same function returned an error) wastes real time and
// bandwidth on a flaky connection. Requests a byte Range starting at the
// existing .part file's size, if any; a server that doesn't honor Range
// (responds 200 instead of 206) gets treated as sending the whole file
// again from byte 0, so the .part is truncated and started over rather
// than getting byte-0 content appended onto existing bytes.
func httpDownloadFile(url, dest string) error {
resp, err := downloadHTTPClient.Get(url)
tmp := dest + ".part"
var offset int64
if fi, err := os.Stat(tmp); err == nil {
offset = fi.Size()
}
req, err := http.NewRequest(http.MethodGet, url, nil)
if err != nil {
return err
}
if offset > 0 {
req.Header.Set("Range", fmt.Sprintf("bytes=%d-", offset))
}
resp, err := downloadHTTPClient.Do(req)
if err != nil {
return err
}
defer resp.Body.Close()
if resp.StatusCode != http.StatusOK {
var out *os.File
switch resp.StatusCode {
case http.StatusOK:
// No partial-content support (or nothing to resume from): the body
// is the whole file from byte 0.
out, err = os.Create(tmp)
case http.StatusPartialContent:
out, err = os.OpenFile(tmp, os.O_WRONLY|os.O_APPEND, 0o644)
case http.StatusRequestedRangeNotSatisfiable:
// Our .part is already >= the real file size - stale or corrupt.
// Discard it; the next retry starts clean with no Range header.
os.Remove(tmp)
return fmt.Errorf("GET %s: range not satisfiable, discarding partial download and retrying from scratch", url)
default:
return fmt.Errorf("GET %s: %s", url, resp.Status)
}
tmp := dest + ".part"
out, err := os.Create(tmp)
if err != nil {
return err
}
if _, err := io.Copy(out, resp.Body); err != nil {
out.Close()
os.Remove(tmp)
// Deliberately not removing tmp here: whatever bytes made it to
// disk are exactly what the next attempt should resume from.
return err
}
if err := out.Close(); err != nil {
os.Remove(tmp)
return err
}
return os.Rename(tmp, dest)
+173
View File
@@ -0,0 +1,173 @@
package download
import (
"fmt"
"net/http"
"net/http/httptest"
"os"
"path/filepath"
"strconv"
"strings"
"testing"
)
func TestHTTPDownloadFileFullDownload(t *testing.T) {
const body = "the quick brown fox jumps over the lazy dog"
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
fmt.Fprint(w, body)
}))
defer srv.Close()
dest := filepath.Join(t.TempDir(), "out")
if err := httpDownloadFile(srv.URL, dest); err != nil {
t.Fatal(err)
}
got, err := os.ReadFile(dest)
if err != nil {
t.Fatal(err)
}
if string(got) != body {
t.Errorf("downloaded content = %q, want %q", got, body)
}
}
// rangeServer serves a fixed body and honors byte-range requests, exactly
// like a real payload host (GitHub Releases, nuget.org, etc.) would.
func rangeServer(body string) *httptest.Server {
return httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
rng := r.Header.Get("Range")
if rng == "" {
w.WriteHeader(http.StatusOK)
fmt.Fprint(w, body)
return
}
var start int
if _, err := fmt.Sscanf(rng, "bytes=%d-", &start); err != nil || start < 0 || start > len(body) {
w.WriteHeader(http.StatusRequestedRangeNotSatisfiable)
return
}
w.Header().Set("Content-Range", "bytes "+strconv.Itoa(start)+"-"+strconv.Itoa(len(body)-1)+"/"+strconv.Itoa(len(body)))
w.WriteHeader(http.StatusPartialContent)
fmt.Fprint(w, body[start:])
}))
}
func TestHTTPDownloadFileResumesFromExistingPart(t *testing.T) {
const body = "the quick brown fox jumps over the lazy dog"
srv := rangeServer(body)
defer srv.Close()
dest := filepath.Join(t.TempDir(), "out")
partial := body[:10]
if err := os.WriteFile(dest+".part", []byte(partial), 0o644); err != nil {
t.Fatal(err)
}
if err := httpDownloadFile(srv.URL, dest); err != nil {
t.Fatal(err)
}
got, err := os.ReadFile(dest)
if err != nil {
t.Fatal(err)
}
if string(got) != body {
t.Errorf("resumed download content = %q, want %q (partial %q should have been continued, not duplicated or lost)", got, body, partial)
}
}
func TestHTTPDownloadFileRestartsWhenServerIgnoresRange(t *testing.T) {
const body = "the quick brown fox jumps over the lazy dog"
// Always answers 200 with the full body, regardless of Range - some
// servers/CDNs genuinely don't support partial content.
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
fmt.Fprint(w, body)
}))
defer srv.Close()
dest := filepath.Join(t.TempDir(), "out")
// A stale/bogus .part that must NOT end up prepended to the real
// content - if httpDownloadFile appended instead of truncating here,
// the result would start with this garbage.
if err := os.WriteFile(dest+".part", []byte("GARBAGE-FROM-A-STALE-ATTEMPT"), 0o644); err != nil {
t.Fatal(err)
}
if err := httpDownloadFile(srv.URL, dest); err != nil {
t.Fatal(err)
}
got, err := os.ReadFile(dest)
if err != nil {
t.Fatal(err)
}
if string(got) != body {
t.Errorf("content = %q, want exactly %q (no leftover garbage prepended)", got, body)
}
}
func TestHTTPDownloadFileKeepsPartOnMidTransferFailure(t *testing.T) {
const fullBody = "0123456789"
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
fmt.Fprint(w, fullBody[:5])
if f, ok := w.(http.Flusher); ok {
f.Flush()
}
// Simulate a dropped connection partway through by closing the
// underlying connection abruptly instead of finishing the body.
hj, ok := w.(http.Hijacker)
if !ok {
return
}
conn, _, err := hj.Hijack()
if err == nil {
conn.Close()
}
}))
defer srv.Close()
dest := filepath.Join(t.TempDir(), "out")
err := httpDownloadFile(srv.URL, dest)
if err == nil {
t.Fatal("expected an error from the truncated connection")
}
partial, err := os.ReadFile(dest + ".part")
if err != nil {
t.Fatalf("expected the .part file with the bytes received so far to survive a failed download: %v", err)
}
if !strings.HasPrefix(fullBody, string(partial)) || len(partial) == 0 {
t.Errorf(".part content = %q, want a non-empty prefix of %q", partial, fullBody)
}
}
func TestHTTPDownloadFileRangeNotSatisfiableDiscardsPart(t *testing.T) {
const body = "short"
srv := rangeServer(body)
defer srv.Close()
dest := filepath.Join(t.TempDir(), "out")
// .part is already longer than the real file - triggers 416 from
// rangeServer's own bounds check.
if err := os.WriteFile(dest+".part", []byte("this partial file is way too long"), 0o644); err != nil {
t.Fatal(err)
}
if err := httpDownloadFile(srv.URL, dest); err == nil {
t.Fatal("expected an error on the first (416) attempt")
}
if _, err := os.Stat(dest + ".part"); !os.IsNotExist(err) {
t.Error("expected the stale .part to be discarded after a 416 response")
}
// The retry (a fresh caller, no Range header since .part is gone) should
// now succeed cleanly.
if err := httpDownloadFile(srv.URL, dest); err != nil {
t.Fatalf("retry after discarding the stale .part failed: %v", err)
}
got, err := os.ReadFile(dest)
if err != nil {
t.Fatal(err)
}
if string(got) != body {
t.Errorf("content = %q, want %q", got, body)
}
}