R语言并行保存绘图失败排查：串行正常并行无输出

问题描述 #

编写的R代码用于生成并保存时序绘图，普通for循环串行执行时可正常保存绘图，但改用doParallel和foreach实现并行执行后，长时间无绘图输出且CPU未被占用，需排查操作错误。

原代码 #

分析函数 #

rm(list=ls())
analysis <- function(time_series,sublist){
    ROIs <- c("RW","RFHM","RFD2","RFD3","RFD4")
    data <- time_series[[sublist]]
    for (subject in 1:length(data$successful_x_direction)){
        subject_name <- paste0("subject_",subject)
        success_x <- data$successful_x_direction[[subject]]
        success_y <- data$successful_y_direction[[subject]]
        fail_x <- data$failure_x_direction[[subject]]
        fail_y <- data$failure_y_direction[[subject]]
        for (success in 1:length(success_x)){
            matrix_x <- success_x[[success]]
            matrix_y <- success_y[[success]]
            matrix_x <- as.data.frame(matrix_x)
            matrix_x$Frame <- seq(1,dim(matrix_x)[1])
            matrix_x <- matrix_x %>% pivot_longer(!Frame,names_to = "ROI",values_to = "Value")
            x_cor <- ggplot(matrix_x,aes(x = Frame, y = Value, colour = ROI)) + geom_line() + ylab("X Coordinates")
            matrix_y <- as.data.frame(matrix_y)
            matrix_y$Frame <- seq(1,dim(matrix_y)[1])
            matrix_y <- matrix_y %>% pivot_longer(!Frame,names_to = "ROI",values_to = "Value")
            y_cor <- ggplot(matrix_y,aes(x = Frame, y = Value, colour = ROI)) + geom_line() +  ylab("Y Coordinates")
            g <- ggarrange(x_cor,y_cor,ncol = 1, nrow = 2)
            g <- annotate_figure(g, top = text_grob("Success", color = "red", face = "bold", size = 14))
           filename = paste0("Success_",success,".png")
           filename = paste("Time Series Plots",sublist,subject_name,filename,sep = "\"")
           ggsave(plot = g, filename = filename)
    }
}

串行执行代码 #

load("time_series.Rdata")
for (i in seq(1,length(time_series))){
    sublist <- names(time_series)[i]
    analysis(time_series,sublist)
}

并行执行代码 #

library(doParallel)
library(foreach)
numCores <- detectCores()-1
cl <- makeCluster(numCores)
registerDoParallel(cl)
random_name <- foreach(i=seq(1,length(time_series))) %dopar% {
    sublist <- names(time_series)[i]
    analysis(time_series,sublist)
}
stopCluster(cl)

错误排查与修正方案 #

1. 并行子进程未加载依赖包 #

analysis函数依赖ggplot2、ggpubr（ggarrange/annotate_figure/text_grob）、tidyr（pivot_longer）等包，但并行子进程不会自动继承主环境的包，必须显式加载。

2. 文件路径拼接语法错误 #

原代码中路径分隔符用了"\"，属于语法错误，应该用"/"或"\\"；同时若目标目录不存在，ggsave会报错，需提前创建目录。

3. 大对象重复传递导致性能瓶颈 #

直接将整个time_series传递给每个并行进程，会引发大量数据复制，甚至导致进程卡住。应仅传递子进程所需的分片数据，或通过clusterExport优化对象传递。

4. 并行环境图形设备兼容问题 #

ggplot在并行环境中可能需要显式指定图形设备，避免设备初始化失败。

修正后的并行代码 #

library(doParallel)
library(foreach)
# 加载主环境依赖包
library(ggplot2)
library(ggpubr)
library(tidyr)

numCores <- detectCores()-1
cl <- makeCluster(numCores)
# 给所有子进程加载必要包
clusterEvalQ(cl, {
    library(ggplot2)
    library(ggpubr)
    library(tidyr)
})
# 传递time_series到子进程环境
clusterExport(cl, "time_series")

registerDoParallel(cl)

# 修正后的分析函数
analysis <- function(time_series,sublist){
    ROIs <- c("RW","RFHM","RFD2","RFD3","RFD4")
    data <- time_series[[sublist]]
    for (subject in 1:length(data$successful_x_direction)){
        subject_name <- paste0("subject_",subject)
        # 递归创建目标目录
        dir_path <- file.path("Time Series Plots", sublist, subject_name)
        if(!dir.exists(dir_path)) dir.create(dir_path, recursive = TRUE)
        
        success_x <- data$successful_x_direction[[subject]]
        success_y <- data$successful_y_direction[[subject]]
        for (success in 1:length(success_x)){
            matrix_x <- success_x[[success]]
            matrix_y <- success_y[[success]]
            matrix_x <- as.data.frame(matrix_x)
            matrix_x$Frame <- seq(1,dim(matrix_x)[1])
            matrix_x <- matrix_x %>% pivot_longer(!Frame,names_to = "ROI",values_to = "Value")
            x_cor <- ggplot(matrix_x,aes(x = Frame, y = Value, colour = ROI)) + geom_line() + ylab("X Coordinates")
            matrix_y <- as.data.frame(matrix_y)
            matrix_y$Frame <- seq(1,dim(matrix_y)[1])
            matrix_y <- matrix_y %>% pivot_longer(!Frame,names_to = "ROI",values_to = "Value")
            y_cor <- ggplot(matrix_y,aes(x = Frame, y = Value, colour = ROI)) + geom_line() +  ylab("Y Coordinates")
            g <- ggarrange(x_cor,y_cor,ncol = 1, nrow = 2)
            g <- annotate_figure(g, top = text_grob("Success", color = "red", face = "bold", size = 14))
            
            filename = paste0("Success_",success,".png")
            filename = file.path(dir_path, filename)
            # 显式指定图形设备
            ggsave(plot = g, filename = filename, device = "png")
        }
    }
}

# 并行执行逻辑
random_name <- foreach(i=seq(1,length(time_series))) %dopar% {
    sublist <- names(time_series)[i]
    analysis(time_series,sublist)
}

stopCluster(cl)

关键修改说明 #

• 给并行子进程显式加载所有依赖包，确保绘图函数可正常调用
• 用file.path生成合法路径，递归创建目标目录避免保存失败
• 优化数据传递逻辑，减少大对象的跨进程重复复制
• 显式指定ggsave的图形设备，适配并行环境的设备初始化规则

内容的提问来源于stack exchange，提问作者Orestis