From: Vladimír Čunát <vladimir.cunat@nic.cz>
Date: Thu, 27 Jul 2017 12:22:17 +0000 (+0200)
Subject: rrcache/predict: tweak *when* an RR is expiring
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=refs%2Fheads%2Fexpiring;p=thirdparty%2Fknot-resolver.git

rrcache/predict: tweak *when* an RR is expiring

1% TTL is just too little when used for refreshing.
Even 50% would mean doing at most double work than necessary,
so let's have this compromise of ~7% TTL.
---

diff --git a/lib/layer/rrcache.c b/lib/layer/rrcache.c
index 2a25c1af2..cd8ba6c2d 100644
--- a/lib/layer/rrcache.c
+++ b/lib/layer/rrcache.c
@@ -44,12 +44,22 @@
 #include "lib/resolve.h"
 
 #define VERBOSE_MSG(qry, fmt...) QRVERBOSE((qry), " rc ",  fmt)
-#define DEFAULT_MINTTL (5) /* Short-time "no data" retention to avoid bursts */
 
-/** Record is expiring if it has less than 1% TTL (or less than 5s) */
+static inline const uint32_t get_min_ttl(const struct kr_cache *cache)
+{
+	/* Short-time "no data" retention to avoid bursts */
+	return MAX(5, cache->ttl_min);
+}
+
+/** Record is expiring if it has roughly less than 7% of original TTL.
+ *
+ * Example expirations: < 5s always, < ~9s for 60s TTL,
+ * 	< 3.8 min. with 1h TTL, < 6.25% TTL asymptotically.
+ * Sync: ../../modules/predict/README.rst
+ */
 static inline bool is_expiring(const knot_rrset_t *rr, uint32_t drift)
 {
-	return 100 * (drift + 5) > 99 * knot_rrset_ttl(rr);
+	return (drift + 5) > knot_rrset_ttl(rr) * 15 / 16;
 }
 
 static int loot_rr(struct kr_cache *cache, knot_pkt_t *pkt, const knot_dname_t *name,
@@ -90,6 +100,7 @@ static int loot_rr(struct kr_cache *cache, knot_pkt_t *pkt, const knot_dname_t *
 
 	if (is_expiring(&cache_rr, drift)) {
 		qry->flags |= QUERY_EXPIRING;
+		/* TODO: also set this when negative cache is used */
 	}
 
 	if ((*flags) & KR_CACHE_FLAG_WCARD_PROOF) {
@@ -372,7 +383,7 @@ static int stash_selected(struct kr_request *req, knot_pkt_t *pkt, map_t *stash,
 		return kr_ok();
 	}
 
-	uint32_t min_ttl = MAX(DEFAULT_MINTTL, req->ctx->cache.ttl_min);
+	uint32_t min_ttl = get_min_ttl(&req->ctx->cache);
 	/* uncached entries are located at the end */
 	for (ssize_t i = arr->len - 1; i >= 0; --i) {
 		ranked_rr_array_entry_t *entry = arr->at[i];
diff --git a/modules/predict/README.rst b/modules/predict/README.rst
index bd5a8dd48..67892b254 100644
--- a/modules/predict/README.rst
+++ b/modules/predict/README.rst
@@ -3,7 +3,7 @@
 Prefetching records
 -------------------
 
-The module refreshes records that are about to expire when they're used (having less than 1% of original TTL).
+The module refreshes records that are about to expire when they're used (having less than 7% of original TTL).
 This improves latency for frequently used records, as they are fetched in advance.
 
 It is also able to learn usage patterns and repetitive queries that the server makes. For example, if