[BUG] `up` pulls an image once per service when several services share it
维护者通常 1 天内回复
评估
这个 Issue 还没有评估数据。
描述
Description
When several services use the same missing image, docker compose up sends one pull request
per service. docker compose pull pulls the image once (fixed for #8768).
With the containerd image store, every one of these pulls also unpacks the layers. On a stack of
21 services sharing one 5.7 GB image, up ran 21 parallel unpacks of the same layers on an
NVIDIA Jetson Thor. containerd logged about 17,600 (*service).Write failed ... ref moby/1/layer-sha256:... locked errors while the pulls waited on each other, and the progress
display stayed at 5.675 GB / 5.683 GB for several minutes while the load average was about 35.
Steps To Reproduce
compose.yaml:
services:
a:
image: alpine:3.20.3
command: ["sleep", "infinity"]
b:
image: alpine:3.20.3
command: ["sleep", "infinity"]
c:
image: alpine:3.20.3
command: ["sleep", "infinity"]
repro.sh (removes alpine:3.20.3 and counts the engine's pull events):
#!/usr/bin/env bash
set -euo pipefail
cd "$(dirname "$0")"
count_pulls() {
docker compose down --timeout 0 >/dev/null 2>&1
docker image rm --force alpine:3.20.3 >/dev/null 2>&1 || true
docker events --filter type=image --filter event=pull --format '{{.Action}} {{.Actor.ID}}' > events.log &
local events_pid=$!
sleep 1
docker compose "$@" >/dev/null 2>&1
sleep 2
kill "$events_pid"
echo "docker compose $*: $(wc -l < events.log) pull(s)"
}
count_pulls pull
count_pulls up --detach
docker compose down --timeout 0 >/dev/null 2>&1
rm -f events.log
Output:
docker compose pull: 1 pull(s)
docker compose up --detach: 3 pull(s)
Expected: up --detach pulls the image once, like pull.
Compose Version
Docker Compose version v5.6.1
Docker Environment
Client and server 29.9.2, Fedora Linux 45 (Workstation Edition), x86_64,
storage driver overlayfs with io.containerd.snapshotter.v1
Anything else?
No response
- 主要语言
- Go
- 星标
- 38.9k
- 派生
- 6k
- 平均合并
- 2 天 22 小时
- 30 天内合并 PR
- 81
环境准备
- 提供 Dockerfile 或 Docker Compose 文件
- 有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue,再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库,在一个分支上完成修改。
- 提交 Pull Request,并在描述里引用这个 Issue 编号。
docker/compose 的其他 Issue
-
[BUG] `docker compose port` prints `:0` (exit 0) for a port the image exposes but the service does not publish可能已有人在做 @ndeloof 于 1 天前认领。 未关闭
难度 2/5 1-3 小时 新手友好度 78/100
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 62/100
docker/compose#14435 · 2 条评论 ·
维护者通常 1 天内回复
-
[BUG] deploy.restart_policy without condition drops the restart policy可能重新可做 @akshat-lakhera 于 25 天前认领,目前没有进行中的 PR。 未关闭
难度 2/5 1-3 小时 新手友好度 82/100
docker/compose#14383 · 3 条评论 ·
维护者通常 1 天内回复
-
[BUG] up: an optional (required: false) service_healthy dependent is not started when a required dependent shares the failing wait可能已有人在做 @ndeloof 于 2 天前认领。 未关闭
难度 4/5 3-5 天 新手友好度 58/100
docker/compose#14507 · 2 条评论 ·
维护者通常 1 天内回复
-
[BUG] Negative replica count in scale flags (docker compose scale / --scale) is accepted without validation可能已有人在做 @arsalanyavari 于 8 天前认领。 未关闭kind/bug status/approved
难度 2/5 1-3 小时 新手友好度 30/100
维护者通常 1 天内回复
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 78/100
IBM-Cloud/terraform-provider-ibm#7115 ·
维护者通常 2 天内回复
-
难度 2/5 1-3 小时 新手友好度 86/100
fluxcd/notification-controller#1441 ·
维护者通常 1 天内回复
-
bug
难度 2/5 1-3 小时 新手友好度 66/100
volcengine/volcengine-go-sdk#163 · 1 条评论 ·
-
难度 2/5 1-3 小时 新手友好度 76/100
OneBusAway/maglev#2145 ·
维护者通常 2 天内回复
-
priority/medium status/ready type/chore
难度 2/5 1-3 小时 新手友好度 84/100
discobox-ai/discobox#1357 ·
维护者通常 1 天内回复