HttpBinaryCacheStore: Retry on transient HTTP errors

This makes us more robust against 500 errors from CloudFront or S3
(assuming the 500 error isn't cached by CloudFront...).

src/libstore/download.cc

@@ -8,6 +8,7 @@
 #include <curl/curl.h>
 
 #include <iostream>
+#include <thread>
 
 
 namespace nix {
@@ -194,7 +195,11 @@ struct CurlDownloader : public Downloader
         if (res != CURLE_OK) {
             Error err =
                 httpStatus == 404 ? NotFound :
-                httpStatus == 403 ? Forbidden : Misc;
+                httpStatus == 403 ? Forbidden :
+                (httpStatus == 408 || httpStatus == 500 || httpStatus == 503
+                 || httpStatus == 504  || httpStatus == 522 || httpStatus == 524
+                 || res == CURLE_COULDNT_RESOLVE_HOST) ? Transient :
+                Misc;
             if (res == CURLE_HTTP_RETURNED_ERROR && httpStatus != -1)
                 throw DownloadError(err, format("unable to download ‘%s’: HTTP error %d")
                     % url % httpStatus);
@@ -210,6 +215,10 @@ struct CurlDownloader : public Downloader
 
     DownloadResult download(string url, const DownloadOptions & options) override
     {
+        size_t attempt = 0;
+
+        while (true) {
+            try {
                 DownloadResult res;
                 if (fetch(resolveUri(url), options)) {
                     res.cached = false;
@@ -218,6 +227,14 @@ struct CurlDownloader : public Downloader
                     res.cached = true;
                 res.etag = etag;
                 return res;
+            } catch (DownloadError & e) {
+                attempt++;
+                if (e.error != Transient || attempt >= options.tries) throw;
+                auto ms = 25 * (1 << (attempt - 1));
+                printMsg(lvlError, format("warning: %s; retrying in %d ms") % e.what() % ms);
+                std::this_thread::sleep_for(std::chrono::milliseconds(ms));
+            }
+        }
     }
 };
 

src/libstore/download.hh

@@ -9,10 +9,11 @@ namespace nix {
 
 struct DownloadOptions
 {
-    string expectedETag;
-    bool verifyTLS{true};
-    enum { yes, no, automatic } showProgress{yes};
-    bool head{false};
+    std::string expectedETag;
+    bool verifyTLS = true;
+    enum { yes, no, automatic } showProgress = yes;
+    bool head = false;
+    size_t tries = 1;
 };
 
 struct DownloadResult
@@ -31,7 +32,7 @@ struct Downloader
     Path downloadCached(ref<Store> store, const string & url, bool unpack,
         const Hash & expectedHash = Hash());
 
-    enum Error { NotFound, Forbidden, Misc };
+    enum Error { NotFound, Forbidden, Misc, Transient };
 };
 
 ref<Downloader> makeDownloader();

src/libstore/http-binary-cache-store.cc

@@ -58,6 +58,7 @@ protected:
             DownloadOptions options;
             options.showProgress = DownloadOptions::no;
             options.head = true;
+            options.tries = 5;
             downloader->download(cacheUri + "/" + path, options);
             return true;
         } catch (DownloadError & e) {
@@ -79,6 +80,7 @@ protected:
         auto downloader(downloaders.get());
         DownloadOptions options;
         options.showProgress = DownloadOptions::no;
+        options.tries = 3;
         try {
             return downloader->download(cacheUri + "/" + path, options).data;
         } catch (DownloadError & e) {