module

Krikri::AptRepositoryCacheRetry

Cache-update retry behavior for the apt_repository plugin, mirroring Ansible's own apt_repository.py update_cache_retries / update_cache_retry_max_delay parameters (defaults 5 / 12, both verified against the locally installed module's argument_spec).

Ansible's loop (apt_repository.py, if update_cache: block): for retry in range(update_cache_retries) around python-apt's Cache().update(), with delay = 2 ** retry + randomize between attempts (randomize = secrets.randbelow(1000) / 1000.0), the delay capped at update_cache_retry_max_delay + randomize when the exponential term outgrows it. Same formula here.

Mixed into AptRepositoryPlugin so the retry loop is callable as a private method, while the regression spec can require this file directly (without firing apt_repository.cr's entry point) and drive the loop through an injected exec_remote proc - the same pattern AptLockRetry established for the apt plugin's own update_cache_retries support.

Constants

DEFAULT_UPDATE_CACHE_RETRIES = 5
DEFAULT_UPDATE_CACHE_RETRY_MAX_DELAY = 12

Class methods

sleeping_seconds(delay : Float64) : Int32

The "Sleeping for N seconds" warning's integer. Ansible does NOT truncate the float delay (%i would give 1.7s -> "1"): both its callers (apt_repository.py:746, apt.py:1426, ansible-core 2.19.11) format int(round(delay)), and Python round is ties-to-even, so 1.7 -> "2", 2.5 -> "2", 3.5 -> "4". Truncating instead changes N whenever the jitter lands in [0.5, 1.0), which changes how repeated warnings dedupe across loop items.

Source

Instance methods

apt_repository_cache_update_with_retry(retries : Int32, max_delay : Int32, exec_remote : Proc(String, NamedTuple(exit_code: Int32, stdout: String, stderr: String)))

Runs apt-get update, retrying on non-zero exit up to retries total attempts (Ansible's range(update_cache_retries) bound) with the exponential-backoff-with-jitter delay between attempts.

Only a non-zero exit code retries: that maps to apt's own fetch failure, the case python-apt reports as FetchFailedException (the only exception Ansible's loop catches). A GPG signature warning leaves apt-get update exiting 0 - that path is handled separately by the plugin's gpg_signature_failure? scan and, like Ansible's non-FetchFailedException exceptions, fails without retrying.

Source
apt_repository_retry_delay(retry : Int32, max_delay : Int32, jitter : Float64 = (Random.rand(1000)) / 1000.0) : Float64

Backoff delay before retry number retry (0-based), matching Ansible's delay = 2 ** retry + randomize capped at update_cache_retry_max_delay + randomize. jitter is injectable so the regression spec can pin the value; defaults to the same 0..1 random fraction Ansible uses.

Source