XPost: linux.debian.maint.dpkg 
 From: guillem@debian.org 
  
 Control: tag -1 unreproducible moreinfo 
  
 Hi! 
  
 On Mon, 2025-10-06 at 21:00:47 +0200, Julian Andres Klode wrote: 
 > On Mon, Oct 06, 2025 at 09:54:12PM +0300, bayram karahan wrote: 
 > > Package: apt 
 > > Version: 3.0.0~alpha3 
 > > Severity: normal 
 > > Tags: trixie performance 
  
 > > On Debian 13 (Trixie, Testing) with the new APT package manager (3.x 
 > > series), package installation performance has significantly regressed 
 > > compared to previous versions. 
 > > 
 > > Installing a simple package used to take around 20€€€30 seconds with 
 older 
 > > APT releases (e.g., 2.6.x in Debian 12), but now the same operation often 
 > > takes 2€€€3 minutes. This slowdown is especially noticeable on slower 
 disks 
 > > or network-based installs. 
 > 
 > What is a simple package? 
  
 Yes, please. 
  
 > > Observed behavior: 
 > > - The €€€Downloading€€€ phase runs at normal speed. 
 > > - The €€€Configuring packages€€€ or €€€Setting up€€€ phases take much 
 longer. 
 > 
 > These steps are inside of dpkg, so I reassign accordingly. 
 > 
 > I do not think there is anything actionable here, and it's likely a 
 > problem with your system. 
  
 What is the actual output from the commands where it seems to be 
 taking too long? It's hard to know what to do without at least that, 
 or a way to reproduce this. 
  
 > > - CPU usage remains low (around 5€€€10%). 
 > > - Disk I/O is minimal. 
 > > - It seems APT or dpkg triggers are spending excessive time in internal 
 > > processing. 
 > 
 > Perhaps your disk is failing? If it's not blocked on CPU then it will be 
 > blocked on I/O, but not necessarily writes but ordering of syncs - you 
 > will not see them associated as writes with the process, I'm not sure 
 > if you see those at all as I/O in monitoring. 
 > 
 > It's plausible dpkg added some extra fsync calls this cycle, but I 
 > can't say for sure. 
  
 To me this seems like some package might be doing something heavy in 
 its maintainer scripts (either configure or triggers). There should be 
 no new fsync() involved there. 
  
 Lacking the necessary information asked above, I'll be closing this 
 report in a bit. 
  
 Thanks, 
 Guillem 
  
 --- SoupGate-Win32 v1.05 
  * Origin: you cannot sedate... all the things you hate (1:229/2)