如何解决shinyapps.io中“HTTP/2 stream 1未正常关闭”错误？

问题现象 #

本地运行正常的R Shiny应用，部署到shinyapps.io后出现连接错误，日志报错如下：

2024-01-24T11:05:27.044583+00:00 shinyapps[2720261]: Warning: Error in curl::curl_fetch_memory: HTTP/2 stream 1 was not closed cleanly before end of the underlying stream
2024-01-24T11:05:27.049958+00:00 shinyapps[2720261]:   15: <Anonymous>
2024-01-24T11:05:27.054170+00:00 shinyapps[2720261]:   13: fn
2024-01-24T11:05:27.059003+00:00 shinyapps[2720261]:    8: retry
2024-01-24T11:05:27.063181+00:00 shinyapps[2720261]:    7: connect$retryingStartServer
2024-01-24T11:05:27.067618+00:00 shinyapps[2720261]:    6: eval
2024-01-24T11:05:27.071758+00:00 shinyapps[2720261]:    5: eval
2024-01-24T11:05:27.075890+00:00 shinyapps[2720261]:    4: eval
2024-01-24T11:05:27.079757+00:00 shinyapps[2720261]:    3: eval
2024-01-24T11:05:27.083979+00:00 shinyapps[2720261]:    2: eval.parent
2024-01-24T11:05:27.088113+00:00 shinyapps[2720261]:    1: local

错误源自以下调用bold包批量获取数据的代码段：

library(bold)
species<-unique(checklist[,1])
x<-length(species)
taxon_total = data.frame()
y <- ceiling(x / 300) + 1
i <- 1
while (i < y){
  ini<-1+(300*(i-1))
  fin <- min(300 * i, x)
  tmp <- bold_seqspec(taxon=species[ini:fin], response = TRUE)
  tt <- paste0(rawToChar(tmp$content, multiple = TRUE), collapse = "")
  Encoding(tt) <- "UTF-8"
  taxa <- utils::read.delim(text = tt, header = TRUE, sep = "\t", stringsAsFactors = FALSE, quote = "")
  taxon_total <- rbind(taxon_total,taxa)
  i = i+1
  }
taxon<-taxon_total

示例物种向量：

species<-c("Zeus faber","Zeus capensis","Zenopsis stabilispinosa","Zenopsis nebulosa","Zenopsis conchifer")

解决方案 #

1. 强制使用HTTP/1.1协议 #

shinyapps.io的网络环境对HTTP/2支持存在兼容性问题，在调用bold_seqspec时指定curl选项强制使用HTTP/1.1：

# 修改bold_seqspec调用，添加curlopts参数
tmp <- bold_seqspec(taxon=species[ini:fin], response = TRUE, 
                    curlopts = list(http_version = 0)) # 0对应HTTP/1.1

2. 添加请求延迟避免API限流 #

bold API存在请求频率限制，短时间内批量请求容易触发限流，在循环末尾添加延迟：

# 循环内每次请求后暂停1秒（可根据需求调整时长）
i = i+1
Sys.sleep(1)

3. 优化数据拼接效率 #

原代码用rbind循环拼接数据框效率极低且易出问题，改用列表存储后一次性合并：

# 初始化空列表替代空数据框
taxon_list <- list()
i <- 1
while (i < y){
  ini<-1+(300*(i-1))
  fin <- min(300 * i, x)
  tmp <- bold_seqspec(taxon=species[ini:fin], response = TRUE, 
                      curlopts = list(http_version = 0))
  tt <- paste0(rawToChar(tmp$content, multiple = TRUE), collapse = "")
  Encoding(tt) <- "UTF-8"
  taxa <- utils::read.delim(text = tt, header = TRUE, sep = "\t", stringsAsFactors = FALSE, quote = "")
  taxon_list[[i]] <- taxa # 将结果存入列表
  i = i+1
  Sys.sleep(1)
}
# 一次性合并所有列表元素
taxon_total <- do.call(rbind, taxon_list)

4. 增加错误捕获机制 #

添加tryCatch避免单个请求失败导致整个应用崩溃，同时记录错误信息：

while (i < y){
  ini<-1+(300*(i-1))
  fin <- min(300 * i, x)
  tryCatch({
    tmp <- bold_seqspec(taxon=species[ini:fin], response = TRUE, 
                        curlopts = list(http_version = 0))
    tt <- paste0(rawToChar(tmp$content, multiple = TRUE), collapse = "")
    Encoding(tt) <- "UTF-8"
    taxa <- utils::read.delim(text = tt, header = TRUE, sep = "\t", stringsAsFactors = FALSE, quote = "")
    taxon_list[[i]] <- taxa
  }, error = function(e){
    # 打印错误日志，方便排查
    message(paste("批次", i, "请求失败：", e$message))
  })
  i = i+1
  Sys.sleep(1)
}

内容的提问来源于stack exchange，提问作者tadeufontes