Relax the threshold for calling getDigestInExclusiveMode().

If the policy goal is to minimize disk seeks, this should be equally good.

RELNOTES: None.
PiperOrigin-RevId: 197611813

1 files changed, 6 insertions, 7 deletions

diff --git a/src/main/java/com/google/devtools/build/lib/actions/cache/DigestUtils.java b/src/main/java/com/google/devtools/build/lib/actions/cache/DigestUtils.java
index d16d1bae89..32dbbcea2b 100644
--- a/src/main/java/com/google/devtools/build/lib/actions/cache/DigestUtils.java
+++ b/src/main/java/com/google/devtools/build/lib/actions/cache/DigestUtils.java
@@ -237,13 +237,12 @@ public class DigestUtils {
 
     // All right, we have neither a fast nor a cached digest. Let's go through the costly process of
     // computing it from the file contents.
-    if (fileSize > 4096 && !MULTI_THREADED_DIGEST.get()) {
-      // We'll have to read file content in order to calculate the digest. In that case
-      // it would be beneficial to serialize those calculations since there is a high
-      // probability that MD5 will be requested for multiple output files simultaneously.
-      // Exception is made for small (<=4K) files since they will not likely to introduce
-      // significant delays (at worst they will result in two extra disk seeks by
-      // interrupting other reads).
+    if (fileSize > 128 * 1024 && !MULTI_THREADED_DIGEST.get()) {
+      // We'll have to read file content in order to calculate the digest.
+      // We avoid overlapping this process for multiple large files, as
+      // seeking back and forth between them will result in an overall loss of
+      // throughput.  Files shorter than the default readahead window of 128KB
+      // on Linux are assumed to be readable in one seek.
       digest = getDigestInExclusiveMode(path);
     } else {
       digest = getDigestInternal(path);