Backport countme bucket calculation fix

doc/dnf5.conf.5.rst

@@ -636,23 +636,50 @@ configuration.
 ``countme``
     :ref:`boolean <boolean-label>`
 
-    Determines whether a special flag should be added to a single, randomly
-    chosen metalink/mirrorlist query each week.
-    This allows the repository owner to estimate the number of systems
-    consuming it, by counting such queries over a week's time, which is much
-    more accurate than just counting unique IP addresses (which is subject to
-    both overcounting and undercounting due to short DHCP leases and NAT,
-    respectively).
-
-    The flag is a simple "countme=N" parameter appended to the metalink and
-    mirrorlist URL, where N is an integer representing the "longevity" bucket
-    this system belongs to.
-    The following 4 buckets are defined, based on how many full weeks have
-    passed since the beginning of the week when this system was installed: 1 =
-    first week, 2 = first month (2-4 weeks), 3 = six months (5-24 weeks) and 4
-    = more than six months (> 24 weeks).
-    This information is meant to help distinguish short-lived installs from
-    long-term ones, and to gather other statistics about system lifecycle.
+    When enabled, one (and only one) HTTP GET request for the metalink file
+    will be selected at random every week to carry a special URL flag.
+
+    This flag allows the repository provider to estimate the number of systems
+    consuming the repository, by counting such requests over a week's time.
+    This method is more accurate than just counting unique IP addresses (which
+    is subject to both overcounting and undercounting due to short DHCP leases
+    and NAT, respectively).
+
+    This is *not* an out-of-band HTTP request made for this purpose alone.
+    Only requests initiated by DNF during normal operation, such as to check
+    for metadata updates, can get this flag.
+
+    The flag is a simple "countme=N" parameter appended to the metalink URL
+    where N is an integer representing the age "bucket" this system belongs to.
+    Four buckets are defined, based on how many full weeks have passed since
+    the installation of a system:
+
+    ======  ===============================
+    bucket  system age
+    ======  ===============================
+    1       first week
+    2       first month (2 - 4 weeks)
+    3       first 6 months (5 - 24 weeks)
+    4       more than 6 months (> 24 weeks)
+    ======  ===============================
+
+    This number is meant to help distinguish short-lived (throwaway) machines
+    from long-term installs and get a better picture of how systems are used
+    over time.
+
+    To determine a system's installation time ("epoch"), the ``machine-id(5)``
+    file's modification time is used as the single source of truth. This file
+    is semantically tied to the system's lifetime as it's typically populated
+    at installation time or during the first boot by an installer tool or init
+    system (such as ``systemd(1)``), respectively, and remains unchanged.
+
+    If the file is empty or missing (such as in containers), the time of the
+    very first request made using the expanded metalink URL (i.e. with any
+    repository variables such as ``$releasever`` substituted) that carried the
+    flag is declared as the epoch.
+
+    If no metalink URL is defined for this repository but a mirrorlist URL is,
+    the latter is used for this purpose instead.
 
     Default: ``False``.
 

libdnf5/repo/repo_downloader.cpp

@@ -31,6 +31,7 @@ along with libdnf.  If not, see <https://www.gnu.org/licenses/>.
 #include <librepo/librepo.h>
 #include <solv/chksum.h>
 #include <solv/util.h>
+#include <sys/stat.h>
 
 #include <filesystem>
 #include <fstream>
@@ -682,6 +683,9 @@ const std::array<const int, 3> COUNTME_BUCKETS = {{2, 5, 25}};
 /// This is to align the time window with an absolute point in time rather
 /// than the last counting event (which could facilitate tracking across
 /// multiple such events).
+///
+/// In the below comments, the window's current position will be referred to
+/// as "this window" for brevity.
 void RepoDownloader::add_countme_flag(LibrepoHandle & handle) {
     auto & logger = *base->get_logger();
 
@@ -711,7 +715,7 @@ void RepoDownloader::add_countme_flag(LibrepoHandle & handle) {
     file_path /= COUNTME_COOKIE;
 
     int ver = COUNTME_VERSION;    // file format version (for future use)
-    time_t epoch = 0;             // position of first-ever counted window
+    time_t epoch = 0;             // position of first observed window
     time_t win = COUNTME_OFFSET;  // position of last counted window
     int budget = -1;              // budget for this window (-1 = generate)
     // TODO(lukash) ideally replace with utils::fs::File (via adding scanf() support?),
@@ -743,8 +747,15 @@ void RepoDownloader::add_countme_flag(LibrepoHandle & handle) {
 
         // Compute the position of this window
         win = now - (delta % COUNTME_WINDOW);
+
+        // Compute the epoch from this system's epoch or, if unknown, declare
+        // this window as the epoch (unless stored in the cookie previously).
+        time_t sysepoch = get_system_epoch();
+        if (sysepoch)
+            epoch = sysepoch - ((sysepoch - COUNTME_OFFSET) % COUNTME_WINDOW);
         if (!epoch)
             epoch = win;
+
         // Window step (0 at epoch)
         int64_t step = (win - epoch) / COUNTME_WINDOW;
 
@@ -753,7 +764,7 @@ void RepoDownloader::add_countme_flag(LibrepoHandle & handle) {
         for (i = 0; i < COUNTME_BUCKETS.size(); ++i)
             if (step < COUNTME_BUCKETS[i])
                 break;
-        uint32_t bucket = i + 1;  // Buckets are indexed from 1
+        uint32_t bucket = i + 1;  // Buckets are numbered from 1
 
         // Set the flag
         std::string flag = "countme=" + std::to_string(bucket);
@@ -781,6 +792,31 @@ std::set<std::string> RepoDownloader::get_optional_metadata() const {
 }
 
 
+/* Returns this system's installation time ("epoch") as a UNIX timestamp.
+ *
+ * Uses the machine-id(5) file's mtime as a good-enough source of truth.  This
+ * file is typically tied to the system's installation or first boot where it's
+ * populated by an installer tool or init system, respectively, and is never
+ * changed afterwards.
+ *
+ * Some systems, such as containers that don't run an init system, may have the
+ * file missing, empty or uninitialized, in which case this function returns 0.
+ */
+time_t RepoDownloader::get_system_epoch() const {
+    std::string filename = "/etc/machine-id";
+    std::string id;
+    struct stat st;
+
+    if (stat(filename.c_str(), &st) != 0 || !st.st_size)
+        return 0;
+    std::ifstream(filename) >> id;
+    if (id == "uninitialized")
+        return 0;
+
+    return st.st_mtime;
+}
+
+
 //void Downloader::download_url(ConfigMain * cfg, const char * url, int fd) {
 //    std::unique_ptr<LrHandle> lr_handle(new_remote_handle(*cfg));
 //    GError * err_p{nullptr};

libdnf5/repo/repo_downloader.hpp

@@ -96,6 +96,7 @@ class RepoDownloader {
 
     std::string get_persistdir() const;
     void add_countme_flag(LibrepoHandle & handle);
+    time_t get_system_epoch() const;
 
     std::set<std::string> get_optional_metadata() const;